Recommender Systems in Tourism: A Systematic Review

1. Introduction to Recommender Systems in Tourism

a. Overview of Recommender Systems

Recommender systems are designed to alleviate information overload by filtering vast amounts of data and providing personalized suggestions tailored to individual users [1]. These systems have become indispensable tools for enhancing user experiences across a multitude of domains, ranging from e-commerce platforms to entertainment services [2]. By analyzing user preferences, behaviors, and contextual information, recommender systems aim to predict and suggest items or content that are most likely to be of interest to a particular user. In the context of e-commerce, these systems might recommend products based on a user's past purchases or browsing history. In the realm of entertainment, they could suggest movies, music, or TV shows based on a user's viewing habits or expressed preferences.

In the tourism sector, recommender systems play a crucial role in assisting users in navigating the overwhelming amount of available data to effectively plan their trips [3], [4]. The sheer volume of information pertaining to destinations, accommodations, activities, and transportation options can be daunting for travelers seeking to create a personalized and fulfilling itinerary. Tourism recommender systems (TRSs) address this challenge by filtering and prioritizing relevant information, enabling users to make informed decisions and optimize their travel experiences. These systems may consider factors such as user preferences (e.g., interests, budget, travel style), contextual information (e.g., location, time of year, weather conditions), and item attributes (e.g., hotel amenities, attraction ratings, activity duration) to generate personalized recommendations.

Furthermore, recommender systems offer a solution to the problem of information overload by providing travel recommendations specifically tailored to tourists [5]. Mobile devices, coupled with recommender system technologies, have become essential tools for mobile users in both travel and business applications. By leveraging these technologies, tourists can access personalized recommendations for attractions, accommodations, and activities, enhancing their overall travel experience. The emerging mobile recommender systems are tailored to mobile device users and promise to substantially enrich tourist experiences, recommending rich multimedia content, context-aware services, and views/ratings of peer users.

b. Importance of Tourism Recommender Systems

The tourism industry holds immense potential for driving economic growth, and recommender systems can play a pivotal role in enhancing the quality of services provided by offering specific and relevant attraction recommendations [6]. By leveraging data mining and machine learning techniques, recommender systems can analyze tourist profiles and location data to suggest attractions that align with their interests and preferences. This not only improves the overall tourist experience but also contributes to the economic growth of the tourism sector by promoting lesser-known attractions and encouraging exploration beyond popular tourist hotspots. Choirul Huda et al. highlight the potential of machine learning and data mining technologies to improve tourism services by providing personalized recommendations for specific attractions based on tourists' location profiles.

Tourism recommender systems (TRSs) are essential for helping users discover interesting destinations and efficiently plan their trips, effectively addressing the pervasive issue of information overload [7], [8]. With the vast amount of information available online, it can be time-consuming and challenging for users to identify the most relevant and appealing travel options. TRSs utilize various recommendation approaches and user profile representation methods to overcome this challenge, providing personalized suggestions that cater to individual preferences and needs. These systems consider a range of factors, including user interests, travel history, budget, and desired travel style, to generate recommendations that align with their unique requirements.

Moreover, personalized tourism experiences are increasingly being designed and implemented through the application of machine learning algorithms within recommender systems [9]. These algorithms analyze user data to generate tailored recommendations for destinations, attractions, accommodations, and activities, taking into account user preferences, past behaviors, and similarities to other users. By leveraging machine learning techniques, TRSs can provide more accurate and relevant recommendations, enhancing the overall travel planning process and ensuring a more satisfying and memorable experience for tourists. M. Badouch and M. Boutaounte emphasize that machine learning algorithms are trained on large datasets to generate personalized recommendations and can continuously improve their effectiveness by incorporating new data and user feedback.

Recommender systems in tourism have the potential to promote sustainability by suggesting eco-friendly activities and directing tourists to less frequented destinations [10]. By integrating sustainability considerations into the recommendation process, TRSs can encourage responsible tourism practices and mitigate the negative impacts of overtourism on popular destinations. These systems can highlight sustainable accommodations, eco-friendly transportation options, and activities that support local communities and preserve natural resources. Furthermore, TRSs can promote lesser-known destinations, diversifying tourist flows and reducing the strain on overcrowded areas.

c. Scope and Objectives of the Review

This review is designed to evaluate the most relevant and accurate techniques currently employed in tourism recommender systems, with the goal of identifying effective strategies for enhancing user experiences and promoting sustainable tourism practices [6]. By systematically analyzing existing literature, this review seeks to provide a comprehensive overview of the current state-of-the-art in tourism recommender systems, highlighting the strengths and limitations of various approaches and identifying opportunities for future research and development. This includes evaluating the algorithms, data sources, and evaluation metrics used in different recommender systems, as well as examining the impact of contextual factors and user preferences on recommendation accuracy and relevance.

Furthermore, this review aims to identify key trends in tourism recommender systems, with a particular focus on the application of machine learning and deep learning techniques for enhancing personalization and recommendation accuracy [11]. By examining recent advancements in these areas, this review seeks to provide insights into the potential of machine learning and deep learning for addressing the challenges and limitations of traditional recommender systems, such as the cold-start problem and data sparsity. This includes exploring the use of various machine learning algorithms, such as collaborative filtering, content-based filtering, and hybrid approaches, as well as examining the application of deep learning architectures, such as neural networks and embeddings, for improving recommendation quality.

The objective is to offer a comprehensive overview of the recommendation approaches utilized in the tourism domain and to propose architectures for hybrid systems that can effectively integrate diverse data sources and recommendation techniques [3], [4]. By examining the strengths and weaknesses of different recommendation approaches, this review seeks to identify opportunities for combining these approaches in hybrid systems that can leverage their complementary strengths to achieve superior performance. This includes exploring the use of hybrid approaches that integrate collaborative filtering, content-based filtering, and knowledge-based reasoning, as well as examining the potential of incorporating contextual information and user feedback into the recommendation process.

This systematic review aims to summarize the state-of-the-art methods and techniques developed by scientists in the field of tourism recommender systems, providing a valuable resource for researchers, practitioners, and policymakers seeking to advance the development and implementation of effective and sustainable tourism recommendation solutions [12]. By synthesizing the existing literature and identifying key trends and challenges, this review seeks to inform future research directions and guide the development of innovative solutions that can enhance the tourist experience, promote sustainable tourism practices, and contribute to the economic growth of the tourism sector. Iustina Ivanova and Mike Wald highlight the importance of standardizing the development of intelligence-based systems in tourism and providing a unified methodology for designed domains.

2. Types of Recommender Systems Used in Tourism

a. Collaborative Filtering

Collaborative filtering is a widely used technique in recommender systems that functions by recommending items based on the preferences of users who exhibit similar tastes or behaviors [13], [14]. This approach leverages the collective intelligence of a user community to identify items that are likely to be of interest to a particular user, based on the preferences of their "neighbors" or similar users. The underlying assumption is that users who have agreed in the past are likely to agree in the future. Balraj Kumar notes that collaborative filtering is a widely deployed technique, and Luis Febre et al. demonstrate its usefulness in tourism.

User-based collaborative filtering, when enhanced with demographic information, can provide highly precise recommendations for touristic sites, catering to the specific needs and preferences of individual travelers [14]. By incorporating demographic data such as age, gender, location, and income level, this approach can refine the recommendation process and generate more relevant and personalized suggestions. For example, a young traveler interested in adventure activities might receive recommendations for hiking trails and extreme sports, while an older traveler seeking cultural experiences might be directed towards historical sites and museums.

Collaborative filtering is further enhanced by integrating demographic information to recommend tourist sites, distinguishing between the needs and preferences of new tourists and registered tourists [14]. New tourists, who have limited interaction history with the system, can be initially profiled based on their demographic information and general preferences, while registered tourists, who have provided more detailed information and interaction data, can receive more personalized recommendations based on their past behaviors and ratings. This adaptive approach ensures that all users receive relevant and valuable recommendations, regardless of their level of engagement with the system.

Geospatial collaborative recommender systems leverage spatial information to identify and recommend interesting travel destinations, taking into account the geographical proximity and spatial relationships between different points of interest [7]. By incorporating spatial data such as location coordinates, distance, and accessibility, these systems can generate recommendations that are tailored to the user's current location and travel plans. For example, a user searching for nearby attractions might receive recommendations for museums, parks, and historical sites within a specified radius, along with information on transportation options and estimated travel times. Zahra Bahramian et al. illustrate the potential of tourism recommender systems as applied to the tourism domain by the implementation of an illustrative geospatial collaborative recommender system using the Foursquare dataset.

b. Content-Based Filtering

Content-based filtering operates by recommending items based on a user's preferences and the features or attributes of the items themselves, creating personalized suggestions [13]. This approach analyzes the characteristics of items that a user has previously liked or interacted with and then recommends similar items based on their content. In the context of tourism, content-based filtering might analyze a user's past travel destinations, preferred activities, and accommodation preferences to recommend similar destinations, activities, and accommodations. Balraj Kumar notes that content-based filtering is a widely deployed technique.

Content-based recommendations often serve as a valuable starting point in tourism recommender systems, providing users with initial suggestions based on their stated preferences and interests [15]. These recommendations can help users explore new destinations, activities, and accommodations that align with their individual tastes, providing a foundation for further exploration and refinement of their travel plans. Vtor Teixeira Camacho and J. Cruz note that content-based recommendations are a good starting point.

Content-based filtering analyzes user data to generate tailored recommendations for destinations, attractions, accommodations, and activities, ensuring that the suggestions align with the user's specific needs and preferences [9]. This approach leverages machine learning algorithms to identify patterns and relationships between user data and item attributes, enabling the system to generate highly personalized and relevant recommendations. For example, a user who has previously expressed an interest in historical sites and cultural experiences might receive recommendations for museums, historical landmarks, and cultural events in their chosen destination.

Content-based systems are particularly effective when combined with demographic data, allowing for a more nuanced and personalized recommendation experience that takes into account the user's individual characteristics and preferences [16]. By incorporating demographic information such as age, gender, location, and income level, these systems can refine the recommendation process and generate suggestions that are more relevant and appealing to the user. For example, a content-based system might recommend family-friendly activities and accommodations to users traveling with children, while suggesting luxury experiences and high-end hotels to affluent travelers. Tamer Uar and Adem Karahoca note that an approach involving demographic and content-based methodologies can eliminate drawbacks.

c. Hybrid Approaches

Hybrid recommender systems represent a powerful and versatile approach that combines multiple recommendation techniques to enhance the overall performance and effectiveness of the system [6], [17]. By integrating different recommendation strategies, hybrid systems can leverage the strengths of each individual approach while mitigating their respective weaknesses, resulting in more accurate, diverse, and personalized recommendations. Choirul Huda et al. note that opportunities are conducted with more usage especially through many types such as content-based, collaborative filtering, demographic, knowledge-based, community-based, hybrid, and Richa Sharma et al. note that hybrid systems combine multiple recommendation techniques to enhance the performance of a single recommendation approach.

Hybrid systems are designed to overcome the limitations of individual methods by strategically fusing two or more techniques, creating a synergistic effect that leads to improved recommendation quality and user satisfaction [18]. For example, a hybrid system might combine collaborative filtering, which relies on user ratings and preferences, with content-based filtering, which analyzes item attributes and features, to generate recommendations that are both relevant and diverse. This approach can help address the cold-start problem, where the system has limited information about new users or items, and improve the accuracy of recommendations for users with diverse or evolving preferences. Bruno Silva and Samuel Arleo note that hybrid approaches fuse two or more techniques in order to overcome the shortcomings of each method.

A hybrid multi-criteria approach can provide destination recommendations that are tailored to user preferences and also take into account a variety of factors, such as budget, travel style, and desired activities [19]. By integrating multiple criteria into the recommendation process, these systems can generate more comprehensive and personalized suggestions that align with the user's specific needs and expectations. For example, a hybrid multi-criteria system might recommend destinations that offer a combination of affordable accommodations, cultural attractions, and outdoor activities, catering to the diverse interests of the user. M S P Maruao and Suharjito Suharjito demonstrate this approach's usefulness.

Hybrid systems often integrate collaborative filtering and content-based filtering to leverage the strengths of both approaches and improve the overall accuracy and relevance of the recommendations [9]. This combination allows the system to generate recommendations that are based on both user preferences and item attributes, resulting in more personalized and diverse suggestions. For example, a hybrid system might use collaborative filtering to identify users with similar tastes and preferences and then use content-based filtering to recommend items that align with those preferences, taking into account the user's individual characteristics and interests. M. Badouch and M. Boutaounte emphasize that hybrid systems that combine both approaches have shown to be effective in producing more accurate recommendations.

3. Context-Aware Recommender Systems in Tourism

a. Definition and Importance of Context-Awareness

Context-aware recommender systems (CARS) are advanced systems that generate recommendations based on a user's current context, taking into account factors such as their location, the time of day, and other relevant environmental or situational variables [20]. By considering the user's context, these systems can provide more relevant and personalized recommendations that are tailored to their immediate needs and preferences. For example, a context-aware system might recommend nearby restaurants during lunchtime or suggest indoor activities on a rainy day. C. V. Sundermann et al. note that these systems have been widely investigated in both academia and industry.

Incorporating contextual information into recommender systems significantly enhances prediction tasks and leads to the delivery of more accurate and appropriate recommendations, improving user satisfaction and engagement [21]. By analyzing contextual data, such as location, time, weather conditions, and social context, these systems can gain a deeper understanding of the user's current needs and preferences, enabling them to generate more relevant and timely suggestions. For example, a system that takes into account the user's location and the current weather conditions might recommend a nearby coffee shop on a cold morning or suggest a scenic walking route on a sunny afternoon. Reshabh Gaude notes that many research were done where it found that beyond user and item profile, incorporating contextual information will enhance the prediction task and providing the more accurate recommendations.

Context-aware systems have become the standard in state-of-the-art recommender systems, reflecting the growing recognition of the importance of contextual information in generating personalized and relevant recommendations [15]. These systems represent a significant advancement over traditional recommender systems, which typically rely solely on user preferences and item attributes, without considering the user's current context. By incorporating contextual factors into the recommendation process, context-aware systems can provide a more holistic and personalized experience, catering to the user's specific needs and preferences in real-time. Vtor Teixeira Camacho and J. Cruz note that RS is context aware as is now the rule in the state-of-the-art for recommender systems.

Context-aware recommender systems enhance traditional recommender systems by leveraging the context of information to generate improved and more relevant recommendations [22]. By considering the user's current situation, these systems can provide suggestions that are tailored to their immediate needs and preferences, leading to increased user satisfaction and engagement. For example, a context-aware system might recommend a nearby hotel based on the user's location, the time of day, and their expressed preferences for accommodation type and amenities. Igor Andr Pegoraro Santana and M. A. Domingues note that Context-Aware Recommender Systems were created, accomplishing state-of-the-art results and improving traditional recommender systems.

b. Types of Contextual Information

Contextual information encompasses a wide range of data points, including user demographics, item attributes, and spatial data, which can be leveraged to enhance the personalization and relevance of recommendations [2], [18]. User demographics, such as age, gender, location, and income level, can provide valuable insights into their preferences and interests. Item attributes, such as category, price, and features, can help the system identify items that align with the user's needs. Spatial data, such as location coordinates and proximity to other points of interest, can be used to generate location-based recommendations. Matthew O. Ayemowa et al. note that auxiliary information, including user demographics, item attributes, and contextual data has shown significant promise in enhancing the performance of recommender systems, and Bruno Silva and Samuel Arleo note that content aware systems provide suggestions based on the situation parameters or conditions that surround the user, while demographic filtering utilizes users demographic characteristics.

Location is a crucial contextual factor, particularly for mobile users who are accessing information and seeking recommendations while on the move, enabling systems to provide geographically relevant suggestions [23]. By leveraging the user's current location, these systems can recommend nearby attractions, restaurants, accommodations, and transportation options, enhancing their travel experience and facilitating their exploration of new destinations. Maede Kiani Sarkaleh notes that mobile be sites museums and demands different people students, tourists ordinary met spite their diverse features preferences.

Temporal context, such as the time of day or the season, can significantly influence recommendations, as user preferences and needs often vary depending on the time of year or the time of day [16]. For example, a user might be interested in different types of activities and attractions during the summer months compared to the winter months, or they might prefer different types of restaurants for lunch versus dinner. By incorporating temporal context into the recommendation process, systems can provide more relevant and timely suggestions that align with the user's current needs and preferences. Tamer Uar and Adem Karahoca note that recommendations will be many factors including users demographic information, past locations favorite seasons.

Emotional states can be leveraged to improve recommendations, as a user's mood and emotional state can influence their preferences and decision-making processes [24]. By analyzing data from sensors and wearables, systems can detect the user's emotional state and generate recommendations that align with their current mood. For example, a user who is feeling stressed might receive recommendations for relaxing activities or calming music, while a user who is feeling energetic might be directed towards more active and stimulating experiences. Luz Santamara-Granados et al. note that recommendation systems have overcome the overload of irrelevant information by considering users preferences and emotional states in fields tourism, health, e-commerce, entertainment.

c. Techniques for Incorporating Context

Ontology-based systems utilize ontologies to group items and facilitate context-aware filtering, enabling the system to reason about the relationships between items and contextual factors [15], [8]. Ontologies provide a structured and formalized representation of knowledge, allowing the system to understand the semantic meaning of items and contextual information. By using ontologies, these systems can generate more accurate and relevant recommendations that take into account the user's current context and preferences. Vtor Teixeira Camacho and J. Cruz note that the presented RS mixes different types of recommenders creating an ensemble which changes on the basis of the RSs maturity, and Zahra Bahramian et al. note that the proposed enhanced using an ontology approach.

Spreading activation techniques can be used to contextualize user preferences and learn user profiles dynamically, adapting the recommendations to the user's evolving needs and interests [8]. These techniques involve assigning weights to different items and contextual factors based on their relevance to the user's preferences, and then propagating these weights through the network to generate recommendations. By dynamically adjusting the weights based on user feedback and behavior, these systems can continuously improve the accuracy and relevance of their recommendations. Zahra Bahramian et al. note that we apply spreading activation technique contextualize user preferences learn profile dynamically according feedback.

Deep learning and embeddings are increasingly being used to improve context-aware recommender systems, enabling the system to learn complex relationships between items, users, and contextual factors [22]. Deep learning models can automatically extract features from raw data, such as text, images, and sensor data, and use these features to generate more accurate and personalized recommendations. Embeddings can be used to represent items and users in a high-dimensional space, capturing their semantic relationships and enabling the system to identify similar items and users based on their contextual profiles. Igor Andr Pegoraro Santana and M. A. Domingues note that A systematic review was conducted to understand how the Deep Learning and Embeddings techniques are being applied to improve Context-Aware Recommender Systems.

Auxiliary information, including user demographics and item attributes, can be leveraged to enhance recommender systems, providing additional context and improving the accuracy and relevance of the recommendations [2]. By incorporating demographic data, such as age, gender, and location, the system can better understand the user's preferences and interests. By analyzing item attributes, such as category, price, and features, the system can identify items that align with the user's needs and preferences. Matthew O. Ayemowa et al. note that This systematic review investigates the impact of incorporating auxiliary information into various types of recommender systems, examining recent advancements, methodologies, datasets, evaluation metrics, and to equally examine its significance on generative artificial intelligence.

4. The Role of Machine Learning in Tourism Recommender Systems

a. Machine Learning Algorithms

Machine learning algorithms play a central role in generating personalized recommendations within tourism recommender systems, enabling these systems to analyze vast amounts of data and identify patterns that reflect individual user preferences and behaviors [9], [17]. By leveraging machine learning techniques, TRSs can provide tailored suggestions for destinations, attractions, accommodations, and activities, enhancing the overall travel planning experience and ensuring that users receive recommendations that align with their specific interests and needs. M. Badouch and M. Boutaounte note that recommender systems that utilize machine learning algorithms are a prominent tool in the design and implementation of personalized tourism experiences, and Richa Sharma et al. note that Over the years, Recommender systems have emerged as a means to provide relevant content to the users, be it in the field of entertainment, social- network, health, education, travel, food or tourism.

Algorithms are meticulously analyzed to solve problems and improve the overall quality of recommendations for massive datasets, ensuring that TRSs can effectively handle the ever-increasing volume of information available online and provide users with relevant and accurate suggestions [25]. By continuously evaluating and refining these algorithms, researchers and developers can optimize the performance of TRSs and enhance their ability to cater to the diverse needs and preferences of travelers. Nayma Khan and Dr. Mohd Haroon note that to solve these kinds of problems, we have found several approaches for making recommendations, including three different types: content-based filtering, collaborative filtering, and hybrid filtering.

Machine learning techniques are employed to evaluate the relevance score of each item-user pair, enabling the system to identify the most suitable recommendations for each individual user based on their unique profile and preferences [26]. By assigning a relevance score to each potential recommendation, TRSs can prioritize the suggestions that are most likely to be of interest to the user, ensuring that they receive the most valuable and personalized recommendations. P. Merinov notes that RSs employ various machine learning techniques to evaluate the relevance score of each item-user pair and then recommend the most relevant items to the corresponding user.

Machine learning algorithms can be trained on large datasets and continuously improve their effectiveness by incorporating new data and user feedback, allowing TRSs to adapt to evolving user preferences and provide increasingly accurate and personalized recommendations over time [9]. By continuously learning from user interactions and feedback, these algorithms can refine their understanding of user preferences and generate more relevant and valuable suggestions, enhancing the overall user experience and ensuring that TRSs remain effective and up-to-date. M. Badouch and M. Boutaounte emphasize that machine learning algorithms can be trained on large datasets to generate personalized recommendations, and can continuously improve their effectiveness by incorporating new data and user feedback.

b. Deep Learning Applications

Deep learning architectures are increasingly being used to generate recommendations and improve context-aware systems, enabling TRSs to capture complex relationships between users, items, and contextual factors and provide more nuanced and personalized suggestions [22]. By leveraging the power of deep learning, these systems can analyze vast amounts of data and identify patterns that would be difficult or impossible to detect using traditional machine learning techniques, leading to more accurate and relevant recommendations. Igor Andr Pegoraro Santana and M. A. Domingues note that There are many approaches to build recommender systems, and two of the most prominent advances in area have been the use of Embeddings to represent the data in the recommender system, and the use of Deep Learning architectures to generate the recommendations to the user.

Deep learning techniques have emerged as a prominent research focus in recommender systems, reflecting the growing recognition of their potential to enhance the accuracy, personalization, and scalability of TRSs [27]. By leveraging deep learning models, researchers and developers can address some of the key challenges facing traditional recommender systems, such as the cold-start problem and data sparsity, and create more effective and user-friendly recommendation solutions. Hongde Zhou et al. note that With the increasing abundance of information resources and the development of deep learning techniques, recommender systems (RSs) based on deep learning have gradually become a research focus.

Deep neural networks (DNNs) have achieved significant advancements in enhancing recommender systems, enabling TRSs to capture complex user preferences and item characteristics and provide more accurate and personalized recommendations [28]. By leveraging the power of DNNs, these systems can analyze vast amounts of data and identify patterns that reflect individual user tastes and behaviors, leading to more relevant and valuable suggestions. Wenqi Fan et al. note that While Deep Neural Networks (DNNs) have achieved significant advancements in enhancing recommender systems, these DNN-based methods still exhibit some limitations, such as inferior capabilities to effectively capture textual side information about users and items, difficulties in generalization to various recommendation scenarios, and reasoning on their predictions, etc.

Deep learning models can effectively incorporate textual side information about users and items, allowing TRSs to leverage the rich semantic content of user reviews, item descriptions, and other textual data sources to generate more informed and personalized recommendations [28]. By analyzing the language used in user reviews and item descriptions, these models can gain a deeper understanding of user preferences and item characteristics, leading to more accurate and relevant suggestions. Wenqi Fan et al. note that LLMs can improve the ability to capture textual side information about users and items.

c. Specific Algorithms and Techniques

Adaptive Neuro-Fuzzy Inference System (ANFIS) and Particle Swarm Optimization (PSO) can be used to optimize multi-criteria recommender systems, enabling TRSs to balance multiple objectives and provide recommendations that satisfy a variety of user needs and preferences [29]. By integrating ANFIS and PSO, these systems can effectively model complex data relationships and generate recommendations that are both accurate and aligned with user expectations. Arash Khosravi and Ahmad Azarnik note that the ANFIS + PSO approach demonstrates superior accuracy and stability, making it highly recommended for future applications that require reliable predictive performance.

Embeddings are used to represent data in recommender systems, allowing TRSs to capture the semantic relationships between users, items, and contextual factors and generate more accurate and personalized recommendations [22]. By mapping users and items to a high-dimensional space, embeddings can capture the subtle nuances of user preferences and item characteristics, enabling the system to identify similar users and items and provide more relevant suggestions. Igor Andr Pegoraro Santana and M. A. Domingues note that Embeddings are used to represent the data in the recommender system.

Case-based reasoning is a technique used to provide recommendations based on previously solved cases, allowing TRSs to leverage past experiences and provide personalized suggestions that are tailored to the user's specific needs and context [30]. By analyzing the characteristics of previously successful recommendations, these systems can identify similar cases and provide corresponding suggestions to new users, enhancing the accuracy and relevance of the recommendations. Tamir Anteneh Alemu et al. note that the provides recommendation previously solved cases new query given by tourist.

Opinion mining and contextual information extraction techniques are used to incorporate user reviews into recommender systems, enabling TRSs to leverage the collective wisdom of the user community and provide more informed and personalized recommendations [20]. By analyzing the sentiment and content of user reviews, these systems can gain valuable insights into the strengths and weaknesses of different items and generate recommendations that are aligned with user preferences and expectations. C. V. Sundermann et al. note that there are increasing efforts to incorporate the rich information embedded in users reviews/texts into the recommender systems.

5. Data Sources and Collection Methods

a. Types of Data Used

User data, including preferences, behavior, demographic profiles, and social network judgments, serves as a cornerstone for developing effective recommender systems, enabling these systems to understand individual user tastes and provide personalized recommendations [9], [2]. User preferences, such as preferred destinations, activities, and accommodation types, provide direct insights into their interests. User behavior, such as past purchases, browsing history, and ratings, reveals their implicit preferences and tendencies. Demographic profiles, including age, gender, location, and income level, offer valuable context for understanding their needs and interests. Social network judgments, such as likes, shares, and comments, provide insights into their social influences and preferences. M. Badouch and M. Boutaounte note that this paper provides a state-of-the-art overview of various types of recommendation systems (RS), including those based on user preferences, behaviors, demographic profiles, and social network judgments, and Matthew O. Ayemowa et al. note that the integration of auxiliary information, including user demographics, item attributes, and contextual data has shown significant promise in enhancing the performance of recommender systems.

Item data, encompassing attributes and features, is crucial for content-based filtering, enabling recommender systems to identify items that align with user preferences based on their inherent characteristics [13]. Item attributes, such as category, price, brand, and features, provide a detailed description of the item's characteristics. Item features, such as keywords, tags, and descriptions, offer additional context for understanding the item's content and purpose. Balraj Kumar notes that content-based systems face certain challenges in deployment cold-start, sparsity, scalability, privacy, etc.

Contextual data, including location and time, plays a vital role in context-aware systems, allowing these systems to tailor recommendations to the user's current situation and provide more relevant and timely suggestions [20]. Location data, such as GPS coordinates and IP address, enables the system to understand the user's current location and recommend nearby attractions, restaurants, and accommodations. Time data, such as the time of day, day of the week, and season, allows the system to adapt recommendations to the user's current schedule and activities. C. V. Sundermann et al. note that Context-aware recommender systems have been widely investigated in both academia and industry because they can make recommendations based on a users current context (e.g., location and time).

Tourist data is generated across various sectors, including hotels, restaurants, transportation, and heritage sites, providing a rich source of information for building and improving tourism recommender systems [3]. Data from hotels, such as occupancy rates, customer reviews, and amenity information, can be used to recommend suitable accommodations to users. Data from restaurants, such as menu items, pricing, and customer ratings, can be used to suggest dining options that align with user preferences. Data from transportation providers, such as flight schedules, train routes, and car rental rates, can be used to plan efficient and convenient travel itineraries. Data from heritage sites, such as historical information, visitor reviews, and accessibility details, can be used to recommend cultural experiences that match user interests. P. Patole notes that With the advancement of the Internet, technology, and communication channels, the generation of tourist data has significantly increased across various sectors such as hotels, restaurants, transportation, heritage sites, tourist events, and activities.

b. Data Collection Techniques

Data is collected from online platforms, including social media and e-commerce sites, providing a wealth of information about user preferences, behaviors, and item characteristics that can be used to build and improve recommender systems [20], [6]. Social media platforms, such as Facebook, Twitter, and Instagram, offer valuable insights into user interests, opinions, and social connections. E-commerce sites, such as Amazon, eBay, and TripAdvisor, provide data on user purchases, browsing history, and ratings. C. V. Sundermann et al. note that the advent of Web 2.0 and the growing popularity of social and e-commerce media sites have encouraged users to naturally write texts describing their assessment of items, and Choirul Huda et al. note that many attractions are detected on several platforms.

Websites and online travel agencies (OTAs) serve as significant sources of tourist data, offering valuable information about destinations, accommodations, activities, and transportation options that can be used to personalize recommendations [3]. Websites, such as travel blogs and destination guides, provide detailed information about tourist attractions, local customs, and travel tips. Online travel agencies (OTAs), such as Expedia, Booking.com, and Airbnb, offer a wide range of travel services, including flights, hotels, and rental cars, and collect data on user bookings and preferences. P. Patole notes that This surge is particularly notable with the rise of Online Travel Agencies (OTAs).

Sensors and wearables can be used to collect data for emotion recognition in tourism recommender systems, enabling these systems to understand user moods and preferences and provide more personalized and responsive recommendations [24]. Sensors, such as cameras and microphones, can capture data about the user's environment, such as lighting, noise levels, and facial expressions. Wearables, such as smartwatches and fitness trackers, can collect data about the user's physiological state, such as heart rate, skin temperature, and activity levels. Luz Santamara-Granados et al. note that the review highlights collection, processing, feature extraction data from sensors wearables detect emotions.

User reviews and ratings are extracted from platforms like TripAdvisor, providing valuable feedback about the quality and appeal of different destinations, accommodations, and activities that can be used to improve the accuracy and relevance of recommender systems [14], [29]. User reviews offer detailed opinions and experiences about different travel options, providing insights into their strengths and weaknesses. User ratings provide a quantitative measure of user satisfaction, allowing the system to rank and prioritize different recommendations. Luis Febre et al. note that the approach is evaluated on a tourist sites dataset extracted from the TripAdvisor platform that contains historical ratings, and demographic information of the tourist, and Arash Khosravi and Ahmad Azarnik note that this research presents a multi-criteria recommender system designed specifically for the travel industry, leveraging a combination of filtering methods applied to TripAdvisor data.

c. Data Preprocessing and Integration

Data preprocessing involves cleaning, transforming, and integrating data from various sources, ensuring that the data is consistent, accurate, and suitable for use in recommender systems [15]. Cleaning the data involves removing errors, inconsistencies, and missing values. Transforming the data involves converting it into a standardized format that can be easily processed by the system. Integrating the data involves combining data from different sources into a unified dataset. Vtor Teixeira Camacho and J. Cruz note that This item classification facilitates the association between user preferences and items, as well as allowing to better classify and group the items being offered, which in turn is particularly useful for context-aware filtering.

Natural language processing (NLP) is used to classify items and associate them with user preferences, enabling the system to understand the semantic meaning of item descriptions and user reviews and generate more relevant recommendations [15]. NLP techniques, such as text mining and sentiment analysis, can be used to extract keywords, topics, and sentiments from text data. These techniques can also be used to identify relationships between items and users based on their textual descriptions and reviews. Vtor Teixeira Camacho and J. Cruz note that the presented RS uses a tourism ontology and natural language processing (NLP) to correctly bin the items to specific item categories and meta categories in the ontology.

Ontologies are used to# Recommender Systems in Tourism: A Systematic Review

1. Introduction to Recommender Systems in Tourism

a. Overview of Recommender Systems

Recommender systems are designed to filter vast amounts of information and provide personalized suggestions to users, making it easier to find relevant content [1]. These systems have become essential tools for enhancing user experiences across various domains, including e-commerce, entertainment, and tourism [2]. In the context of tourism, recommender systems play a crucial role in helping users navigate the overwhelming amount of available data to plan their trips effectively [3], [4]. By analyzing user preferences and providing tailored recommendations, these systems reduce information overload and significantly improve the overall travel planning experience [5].

Recommender systems operate by employing various techniques to understand user preferences and match them with relevant items or services. These techniques include collaborative filtering, content-based filtering, and hybrid approaches, each with its own strengths and limitations. Collaborative filtering relies on the preferences of similar users to make recommendations, while content-based filtering focuses on the attributes of items and user profiles. Hybrid approaches combine these methods to provide more accurate and diverse recommendations. The primary goal of these systems is to provide users with personalized and relevant suggestions, thereby enhancing their decision-making process and overall satisfaction.

In the realm of e-commerce, recommender systems suggest products that a user might be interested in purchasing, based on their past browsing history, purchase behavior, and demographic information. Similarly, in the entertainment industry, these systems recommend movies, music, or TV shows that align with a user's taste. The underlying principle remains the same: to filter the vast amount of available information and present users with options that are most likely to be of interest to them. The success of these systems hinges on their ability to accurately predict user preferences and provide recommendations that are both relevant and diverse.

b. Importance of Tourism Recommender Systems

The tourism industry has emerged as a significant sector with the potential to drive economic growth, and recommender systems can play a vital role in improving its services by providing specific attraction recommendations to tourists [6]. Tourism recommender systems (TRSs) are particularly important because they address the information overload problem by helping users find interesting destinations and plan their trips more efficiently [7], [8]. These systems leverage various data sources and algorithms to provide personalized tourism experiences, utilizing machine learning algorithms to design and implement these experiences effectively [9]. Furthermore, recommender systems in tourism can promote sustainability by suggesting sustainable activities and less popular destinations, contributing to responsible tourism practices [10].

TRSs offer numerous benefits to both tourists and the tourism industry. For tourists, these systems simplify the trip planning process by providing personalized recommendations that align with their preferences and interests. This can save time and effort, allowing tourists to focus on enjoying their travel experiences. For the tourism industry, TRSs can increase revenue by promoting attractions, accommodations, and activities that might otherwise be overlooked. By providing targeted recommendations, these systems can also help to distribute tourism more evenly, reducing overcrowding in popular destinations and promoting lesser-known areas.

Moreover, TRSs can enhance the overall quality of tourism experiences by providing users with access to relevant and up-to-date information. This can include details about attractions, transportation options, local events, and cultural insights. By providing this information in a personalized and accessible format, TRSs can help tourists make informed decisions and have more enriching travel experiences. The integration of context-aware features, such as location-based recommendations and real-time updates, can further enhance the value of these systems.

c. Scope and Objectives of the Review

This systematic review aims to evaluate the most relevant and accurate techniques used in tourism recommender systems, providing a comprehensive overview of the field [6]. The review identifies key trends in tourism recommender systems, including the increasing use of machine learning and deep learning techniques to improve recommendation accuracy and personalization [11]. The primary objective is to provide an overview of the various recommendation approaches employed in the tourism domain and to propose architectures for hybrid systems that can overcome the limitations of individual methods [3], [4]. By summarizing state-of-the-art methods and techniques developed by scientists in the field, this review serves as a valuable resource for researchers and practitioners seeking to develop effective tourism recommender systems [12].

The scope of this review encompasses a wide range of topics related to tourism recommender systems, including collaborative filtering, content-based filtering, hybrid approaches, context-aware systems, and the role of machine learning. The review also examines the various data sources and collection methods used in these systems, as well as the evaluation metrics and methods used to assess their performance. By providing a comprehensive overview of these topics, this review aims to provide readers with a solid understanding of the current state of the field and the key challenges and opportunities that lie ahead.

In addition to providing a broad overview of the field, this review also delves into specific techniques and algorithms used in tourism recommender systems. This includes detailed discussions of collaborative filtering algorithms, content-based filtering techniques, and hybrid approaches that combine these methods. The review also examines the use of machine learning algorithms, such as deep learning, to improve recommendation accuracy and personalization. By providing this level of detail, the review aims to provide readers with the knowledge and tools they need to develop effective tourism recommender systems for a variety of applications.

2. Types of Recommender Systems Used in Tourism

a. Collaborative Filtering

Collaborative filtering is a widely used technique in recommender systems that recommends items based on the preferences of users with similar tastes [13], [14]. This approach leverages the collective intelligence of users to provide personalized recommendations. User-based collaborative filtering, enhanced with demographic information, can provide more precise recommendations for touristic sites, tailoring suggestions to specific user characteristics [14]. By considering both historical ratings and demographic information, collaborative filtering can effectively recommend tourist sites, particularly for new and registered tourist types [14]. Furthermore, geospatial collaborative recommender systems utilize spatial information to identify interesting travel destinations, integrating geographical context into the recommendation process [7].

Collaborative filtering algorithms typically operate in two main phases: neighborhood formation and recommendation generation. In the neighborhood formation phase, the algorithm identifies users who have similar preferences to the target user. This is typically done by calculating a similarity score between users based on their past ratings or interactions with items. In the recommendation generation phase, the algorithm uses the preferences of the target user's neighbors to predict the user's interest in items they have not yet interacted with. The algorithm then recommends the items with the highest predicted interest scores.

There are several variations of collaborative filtering algorithms, including user-based, item-based, and model-based approaches. User-based collaborative filtering, as mentioned earlier, focuses on finding users with similar preferences. Item-based collaborative filtering, on the other hand, focuses on finding items that are similar to those that the target user has liked in the past. Model-based collaborative filtering uses machine learning techniques to build a model of user preferences and item characteristics, which is then used to generate recommendations. Each of these approaches has its own strengths and weaknesses, and the choice of algorithm depends on the specific characteristics of the dataset and the application.

b. Content-Based Filtering

Content-based filtering is another popular technique used in recommender systems, which recommends items based on the similarity between user preferences and item features [13]. This approach relies on understanding the characteristics of items and matching them with the preferences of users. Content-based recommendations are often used as a starting point in tourism recommender systems, providing initial suggestions based on user profiles and item descriptions [15]. By analyzing user data, content-based filtering generates recommendations for destinations, attractions, accommodations, and activities that align with user preferences and past behaviors [9]. Furthermore, content-based systems are particularly useful when combined with demographic data, allowing for more personalized and accurate recommendations [16].

Content-based filtering algorithms typically operate by first creating a profile for each user and each item. The user profile represents the user's preferences, while the item profile represents the characteristics of the item. These profiles are typically created using various techniques, such as keyword extraction, topic modeling, and sentiment analysis. Once the profiles are created, the algorithm calculates a similarity score between the user profile and the item profile. The algorithm then recommends the items with the highest similarity scores.

One of the key advantages of content-based filtering is that it can provide recommendations for new items that have not yet been rated by other users. This is because the algorithm relies on the characteristics of the item, rather than the ratings of other users. Another advantage is that it can provide personalized recommendations that are tailored to the specific preferences of each user. However, content-based filtering also has some limitations. One limitation is that it can be difficult to create accurate profiles for users and items, particularly when dealing with complex or unstructured data. Another limitation is that it can be difficult to provide diverse recommendations, as the algorithm tends to recommend items that are similar to those that the user has liked in the past.

c. Hybrid Approaches

Hybrid recommender systems combine multiple recommendation techniques to enhance overall performance and overcome the limitations of individual methods [6], [17]. By fusing two or more techniques, hybrid systems can address the shortcomings of each individual approach, providing more robust and accurate recommendations [18]. For example, a hybrid multi-criteria approach can provide destination recommendations that align with user preferences by integrating various methods such as content-based filtering, collaborative filtering, and demographic information [19]. These systems often integrate collaborative filtering and content-based filtering to improve recommendation accuracy and personalization, leveraging the strengths of both approaches [9].

Hybrid recommender systems can be implemented in various ways, including weighted averaging, switching, and feature combination. In weighted averaging, the recommendations from different techniques are combined using a weighted average, where the weights are determined based on the performance of each technique. In switching, the system selects the most appropriate technique to use based on the specific characteristics of the user or the item. In feature combination, the features from different techniques are combined to create a more comprehensive representation of the user or the item.

One of the key advantages of hybrid recommender systems is that they can provide more accurate and diverse recommendations than individual techniques. By combining the strengths of different approaches, hybrid systems can overcome the limitations of each individual method. For example, a hybrid system that combines collaborative filtering and content-based filtering can provide recommendations for both new and existing items, as well as personalized recommendations that are tailored to the specific preferences of each user. However, hybrid recommender systems also have some challenges. One challenge is that they can be more complex to implement and maintain than individual techniques. Another challenge is that it can be difficult to determine the optimal way to combine the different techniques.

3. Context-Aware Recommender Systems in Tourism

a. Definition and Importance of Context-Awareness

Context-aware recommender systems enhance the recommendation process by making suggestions based on a user's current context, such as their location, time, and social surroundings [20]. By incorporating contextual information, these systems can enhance prediction tasks and provide more accurate and relevant recommendations [21]. Context-aware systems are now considered the standard in state-of-the-art recommender systems, reflecting the growing recognition of the importance of contextual factors in shaping user preferences [15]. These systems improve upon traditional recommender systems by leveraging the context of information to provide more personalized and timely suggestions [22].

The importance of context-awareness in recommender systems stems from the fact that user preferences and behaviors are often influenced by the context in which they are making decisions. For example, a user's preference for a particular restaurant may depend on the time of day, the location of the restaurant, and the social context (e.g., whether they are dining alone or with friends). By incorporating these contextual factors into the recommendation process, context-aware recommender systems can provide more relevant and personalized suggestions.

There are several benefits to using context-aware recommender systems in tourism. One benefit is that they can provide recommendations that are tailored to the specific needs and preferences of the user at a particular point in time. For example, a context-aware system might recommend a nearby attraction that is open and has good reviews, based on the user's current location and the time of day. Another benefit is that they can help users discover new and interesting places that they might not otherwise have considered. By incorporating contextual factors, these systems can provide recommendations that are both relevant and surprising, enhancing the overall travel experience.

b. Types of Contextual Information

Contextual information encompasses a wide range of factors, including user demographics, item attributes, and spatial data, which can be used to enhance recommender systems [2], [18]. Location is a particularly crucial contextual factor, especially for mobile users who are accessing information on the go and seeking nearby recommendations [23]. Temporal context, such as the time of day or season, can also significantly influence recommendations, as user preferences often vary depending on the time of year or the time of day [16]. Additionally, emotional states can be used to improve recommendations, tailoring suggestions to match the user's mood and emotional needs [24].

User demographics, such as age, gender, and income, can provide valuable insights into user preferences and behaviors. For example, a younger user might be more interested in adventurous activities, while an older user might prefer more relaxing and cultural experiences. Item attributes, such as price, rating, and category, can also be used to filter and rank recommendations. For example, a user might be more interested in high-rated restaurants or affordable accommodations.

Spatial data, such as the location of attractions, restaurants, and hotels, is particularly important in tourism recommender systems. By incorporating spatial data, these systems can provide recommendations that are tailored to the user's current location and the surrounding area. Temporal context, such as the time of day or season, can also play a significant role in shaping user preferences. For example, a user might be more interested in outdoor activities during the summer months or indoor activities during the winter months. Finally, emotional states can provide valuable insights into user needs and preferences. For example, a user who is feeling stressed might be more interested in relaxing activities, while a user who is feeling adventurous might be more interested in thrilling experiences.

c. Techniques for Incorporating Context

Various techniques are used to incorporate context into tourism recommender systems, enhancing their ability to provide personalized and relevant suggestions. Ontology-based systems utilize ontologies to group items and facilitate context-aware filtering, providing a structured approach to managing and utilizing contextual information [15], [8]. Spreading activation techniques can contextualize user preferences and learn user profiles dynamically, adapting to changing user needs and behaviors [8]. Deep learning and embeddings are increasingly used to improve context-aware recommender systems, leveraging the power of machine learning to model complex relationships between users, items, and context [22]. Additionally, auxiliary information, including user demographics and item attributes, can be integrated to enhance the performance and accuracy of recommender systems [2].

Ontology-based systems use ontologies to represent knowledge about users, items, and context. Ontologies provide a structured and formal way to represent relationships between different concepts, allowing the system to reason about user preferences and make more informed recommendations. Spreading activation techniques are used to propagate information through the ontology, activating relevant concepts based on user preferences and context. This allows the system to identify items that are relevant to the user's current needs and interests.

Deep learning and embeddings are used to learn representations of users, items, and context. These representations can capture complex relationships between different factors, allowing the system to make more accurate and personalized recommendations. Deep learning models can be trained on large datasets of user interactions and contextual information, learning to predict user preferences based on a variety of factors. Auxiliary information, such as user demographics and item attributes, can be used to further enhance the performance of these models. By incorporating these techniques, context-aware recommender systems can provide more relevant and personalized recommendations, enhancing the overall travel experience for users.

4. The Role of Machine Learning in Tourism Recommender Systems

a. Machine Learning Algorithms

Machine learning algorithms play a pivotal role in tourism recommender systems, enabling the generation of personalized recommendations based on user data [9], [17]. Algorithms such as collaborative filtering and content-based filtering are employed to analyze user preferences and behaviors, providing tailored suggestions for destinations, attractions, and activities. These algorithms are continuously analyzed to solve problems and improve the quality of recommendations for massive datasets, ensuring that the system remains effective and up-to-date [25]. Machine learning techniques are also used to evaluate the relevance score of each item-user pair, allowing the system to rank recommendations based on their predicted relevance to the user [26]. Furthermore, machine learning algorithms can be trained on large datasets and continuously improve their effectiveness by incorporating new data and user feedback, adapting to changing user preferences and trends [9].

Machine learning algorithms can be broadly classified into supervised, unsupervised, and reinforcement learning techniques. Supervised learning algorithms are trained on labeled data, where the input features and the desired output are known. These algorithms are used to predict user preferences based on historical data, such as past ratings or purchase behavior. Unsupervised learning algorithms, on the other hand, are trained on unlabeled data, where the desired output is not known. These algorithms are used to discover patterns and relationships in the data, such as identifying clusters of users with similar preferences. Reinforcement learning algorithms learn by interacting with the environment and receiving feedback in the form of rewards or penalties. These algorithms are used to optimize the recommendation process, learning to provide recommendations that maximize user engagement and satisfaction.

The choice of machine learning algorithm depends on the specific characteristics of the dataset and the application. For example, collaborative filtering algorithms are well-suited for datasets with a large number of users and items, while content-based filtering algorithms are well-suited for datasets with rich item descriptions. Hybrid approaches that combine multiple machine learning algorithms can often provide more accurate and diverse recommendations than individual techniques. By leveraging the power of machine learning, tourism recommender systems can provide personalized and relevant suggestions that enhance the overall travel experience for users.

b. Deep Learning Applications

Deep learning architectures are increasingly used in tourism recommender systems to generate recommendations and improve context-aware systems, leveraging their ability to model complex relationships between users, items, and context [22]. These techniques have become a significant research focus in recommender systems, reflecting their potential to enhance recommendation accuracy and personalization [27]. Deep neural networks (DNNs) have achieved significant advancements in enhancing recommender systems, providing more sophisticated and nuanced recommendations [28]. Furthermore, deep learning models can incorporate textual side information about users and items, allowing the system to leverage unstructured data sources such as user reviews and social media posts [28].

Deep learning models can be used to learn representations of users, items, and context. These representations can capture complex relationships between different factors, allowing the system to make more accurate and personalized recommendations. For example, deep learning models can be used to learn embeddings of users and items, where each embedding represents the user's or item's characteristics in a low-dimensional space. These embeddings can then be used to calculate similarity scores between users and items, providing a basis for generating recommendations.

Deep learning models can also be used to model the sequential behavior of users. For example, recurrent neural networks (RNNs) can be used to model the sequence of items that a user has interacted with in the past, allowing the system to predict the user's next item of interest. This is particularly useful in tourism recommender systems, where the user's travel plans often involve a sequence of destinations, attractions, and activities. By leveraging the power of deep learning, tourism recommender systems can provide more personalized and relevant suggestions that enhance the overall travel experience for users.

c. Specific Algorithms and Techniques

Specific algorithms and techniques are employed within machine learning to optimize tourism recommender systems, enhancing their precision and effectiveness. Adaptive Neuro-Fuzzy Inference System (ANFIS) and Particle Swarm Optimization (PSO) can optimize multi-criteria recommender systems, improving their ability to consider multiple factors in the recommendation process [29]. Embeddings are used to represent data in recommender systems and improve overall performance, allowing for more efficient and accurate calculations of similarity and relevance [22]. Case-based reasoning is used to provide recommendations based on previously solved cases, leveraging past experiences to guide current suggestions [30]. Furthermore, opinion mining and contextual information extraction techniques are used to incorporate user reviews into recommender systems, providing valuable insights into user preferences and item characteristics [20].

ANFIS is a hybrid technique that combines the strengths of fuzzy logic and neural networks. It can be used to model complex relationships between multiple criteria, such as user preferences, item attributes, and contextual factors. PSO is a population-based optimization algorithm that can be used to optimize the parameters of the ANFIS model, improving its accuracy and performance. Embeddings are used to represent users, items, and context in a low-dimensional space. These embeddings can be learned using various techniques, such as matrix factorization, deep learning, and word embeddings. Case-based reasoning is a technique that uses past experiences to guide current decision-making. In the context of tourism recommender systems, case-based reasoning can be used to recommend destinations, attractions, and activities based on the experiences of other users with similar preferences.

Opinion mining and contextual information extraction techniques are used to extract valuable information from user reviews. Opinion mining, also known as sentiment analysis, is used to determine the overall sentiment expressed in a review, such as positive, negative, or neutral. Contextual information extraction is used to identify specific aspects of the item that are mentioned in the review, such as the food quality, service, or atmosphere. By incorporating these techniques, tourism recommender systems can provide more personalized and relevant suggestions that are based on the opinions and experiences of other users.

5. Data Sources and Collection Methods

a. Types of Data Used

Various types of data are utilized in tourism recommender systems to create personalized and effective recommendations. User data, including preferences, behavior, demographic profiles, and social network judgments, is essential for understanding user needs and tailoring suggestions accordingly [9], [2]. Item data, encompassing attributes and features of tourism products such as hotels, attractions, and activities, is crucial for content-based filtering and matching items to user preferences [13]. Contextual data, including location, time, and social context, is used in context-aware systems to provide recommendations that are relevant to the user's current situation [20]. Additionally, tourist data generated across various sectors, including hotels, restaurants, transportation, and heritage sites, provides a comprehensive view of the tourism landscape and enables more accurate recommendations [3].

User data can be collected through various means, such as explicit ratings, implicit feedback, and demographic information. Explicit ratings involve users providing explicit ratings or reviews for items they have interacted with. Implicit feedback involves tracking user behavior, such as clicks, purchases, and browsing history. Demographic information can be collected through user registration forms or social media profiles. Item data can be collected from various sources, such as online travel agencies, tourism websites, and item descriptions. This data typically includes information about the item's attributes, such as price, rating, category, and location.

Contextual data can be collected from various sources, such as GPS sensors, mobile devices, and social media platforms. This data can include information about the user's location, time, and social context. Tourist data can be collected from various sectors, such as hotels, restaurants, transportation providers, and heritage sites. This data can include information about bookings, transactions, and customer feedback. By integrating these various types of data, tourism recommender systems can provide more personalized and relevant suggestions that enhance the overall travel experience for users.

b. Data Collection Techniques

Data collection techniques in tourism recommender systems are diverse, encompassing online platforms, websites, and specialized tools to gather comprehensive user and item information. Data is often collected from online platforms, including social media and e-commerce sites, to understand user preferences and behaviors [20], [6]. Websites and online travel agencies (OTAs) serve as significant sources of tourist data, providing information on destinations, accommodations, and activities [3]. Sensors and wearables can be used to collect data for emotion recognition in tourism recommender systems, adding a layer of personalization based on user emotional states [24]. Additionally, user reviews and ratings are extracted from platforms like TripAdvisor, providing valuable insights into user experiences and item quality [14], [29].

Data collection from social media platforms involves scraping publicly available data, such as user profiles, posts, and comments. This data can be used to infer user preferences, interests, and social connections. Data collection from e-commerce sites involves tracking user behavior, such as clicks, purchases, and browsing history. This data can be used to understand user preferences and identify items that are likely to be of interest to the user. Data collection from websites and OTAs involves extracting information about destinations, accommodations, and activities. This data can be used to create item profiles and match items to user preferences.

Data collection from sensors and wearables involves tracking user physiological signals, such as heart rate, skin conductance, and facial expressions. This data can be used to infer user emotional states and tailor recommendations accordingly. Data collection from user reviews and ratings involves extracting information about user experiences and item quality. This data can be used to create item profiles and identify items that are likely to be of interest to the user. By integrating these various data collection techniques, tourism recommender systems can gather comprehensive user and item information, enabling more personalized and effective recommendations.

c. Data Preprocessing and Integration

Data preprocessing and integration are critical steps in building effective tourism recommender systems, ensuring that the collected data is clean, consistent, and ready for analysis. Data preprocessing involves cleaning, transforming, and integrating data from various sources, addressing issues such as missing values, inconsistencies, and noise [15]. Natural language processing (NLP) is used to classify items and associate them with user preferences, extracting valuable information from unstructured text data [15]. Ontologies are used to structure and organize data, facilitating the association between user preferences and items by providing a formal representation of knowledge [15]. Furthermore, embeddings are used to represent data in a format suitable for machine learning algorithms, allowing for efficient and accurate calculations of similarity and relevance [22].

Data cleaning involves removing or correcting errors and inconsistencies in the data. This can include handling missing values, correcting typos, and removing duplicate entries. Data transformation involves converting the data into a format that is suitable for analysis. This can include scaling numerical data, encoding categorical data, and normalizing text data. Data integration involves combining data from various sources into a single, unified dataset. This can involve resolving schema conflicts, handling data inconsistencies, and ensuring data quality.

Natural language processing (NLP) techniques are used to extract valuable information from unstructured text data, such as user reviews and item descriptions. This can include sentiment analysis, topic modeling, and keyword extraction. Ontologies are used to represent knowledge about users, items, and context. Ontologies provide a structured and formal way to represent relationships between different concepts, allowing the system to reason about user preferences and make more informed recommendations. Embeddings are used to represent users, items, and context in a low-dimensional space. These embeddings can capture complex relationships between different factors, allowing the system to make more accurate and personalized recommendations. By performing these data preprocessing and integration steps, tourism recommender systems can ensure that the data is clean, consistent, and ready for analysis, enabling more effective and accurate recommendations.

6. Evaluation Metrics and Methods

a. Common Evaluation Metrics

Various evaluation metrics are used to assess the performance and effectiveness of tourism recommender systems, providing insights into their accuracy, relevance, and overall quality. Precision, recall, and F1-score are commonly used to evaluate the quality of recommendations, measuring the accuracy and completeness of the system's suggestions [14], [19]. Root Mean Squared Error (RMSE) is used to evaluate prediction accuracy, quantifying the difference between predicted and actual user ratings [14], [19]. Normalized Discounted Cumulative Gain (NDCG) is used to evaluate the ranking quality of recommendations, assessing the system's ability to present relevant items at the top of the list [31]. However, it's important to note that beyond-accuracy qualities are rarely assessed, highlighting a need for more diverse evaluation metrics that consider factors such as novelty, diversity, and serendipity [31].

Precision measures the proportion of recommended items that are relevant to the user, while recall measures the proportion of relevant items that are recommended by the system. The F1-score is the harmonic mean of precision and recall, providing a balanced measure of the system's accuracy and completeness. RMSE measures the average difference between predicted and actual user ratings, providing a measure of the system's prediction accuracy. NDCG measures the ranking quality of recommendations, taking into account the position of relevant items in the list. A higher NDCG score indicates that the system is better at presenting relevant items at the top of the list.

In addition to these common evaluation metrics, there are several other metrics that can be used to assess the performance of tourism recommender systems. These include coverage, which measures the proportion of items that are recommended by the system; diversity, which measures the variety of items that are recommended by the system; novelty, which measures the proportion of recommended items that are new to the user; and serendipity, which measures the proportion of recommended items that are surprising and interesting to the user. By considering these various evaluation metrics, researchers and practitioners can gain a more comprehensive understanding of the strengths and weaknesses of tourism recommender systems.

b. Offline vs. Online Evaluation

Evaluation methods for tourism recommender systems can be broadly categorized into offline and online evaluations, each offering different advantages and insights. Offline experimentation is the predominant experiment type in research on recommender systems, allowing for controlled testing and analysis of different algorithms and techniques [31]. Online evaluations are primarily used in combination with other experimentation methods, providing real-world feedback on system performance and user satisfaction [31]. User studies are conducted to evaluate different aspects of recommender systems from the users' perspective, gathering qualitative and quantitative data on user experience and preferences [32]. Acceptance testing is used to measure the easiness of use, efficiency, and correctness of recommendations, ensuring that the system meets the needs and expectations of its users [30].

Offline evaluations involve using historical data to simulate user interactions and assess the performance of the recommender system. This allows researchers to test different algorithms and techniques in a controlled environment, without affecting real users. Online evaluations involve deploying the recommender system in a real-world setting and tracking user behavior. This provides valuable feedback on the system's performance and user satisfaction. User studies involve gathering qualitative and quantitative data from users about their experience with the recommender system. This can include surveys, interviews, and focus groups. Acceptance testing involves evaluating the system against a set of predefined criteria to ensure that it meets the needs and expectations of its users.

The choice between offline and online evaluation depends on the specific goals and resources of the evaluation. Offline evaluations are typically less expensive and time-consuming than online evaluations, but they may not accurately reflect real-world user behavior. Online evaluations provide more realistic feedback, but they can be more expensive and time-consuming to conduct. User studies can provide valuable insights into user experience and preferences, but they can be difficult to generalize to the entire user population. Acceptance testing provides a structured approach to evaluating the system, but it may not capture all of the important aspects of user satisfaction. By combining these various evaluation methods, researchers and practitioners can gain a more comprehensive understanding of the strengths and weaknesses of tourism recommender systems.

c. Challenges in Evaluation

Evaluating tourism recommender systems presents several challenges, including data limitations, the need for diverse evaluation metrics, and the difficulty of capturing real-world user behavior. Few datasets are widely used, and many datasets are used in only a few papers each, limiting the ability to compare results across different studies and hindering the development of standardized benchmarks [31]. There is a need for standardization of non-standard evaluation measures, ensuring that different studies use consistent metrics and methods to assess system performance [13]. Current evaluation approaches have significant limitations and must be addressed in future studies, including the need to consider factors such as novelty, diversity, and serendipity [33]. Furthermore, evaluating the impact of sustainability and fairness in tourism recommender systems requires new methodologies that consider the social and environmental consequences of recommendations [10].

One of the key challenges in evaluating tourism recommender systems is the lack of widely used datasets. This makes it difficult to compare results across different studies and hinders the development of standardized benchmarks. There is a need for more publicly available datasets that are representative of real-world tourism scenarios. Another challenge is the lack of standardized evaluation metrics. Many studies use different metrics to assess system performance, making it difficult to compare results across different studies. There is a need for more standardized evaluation metrics that consider factors such as accuracy, relevance, diversity, novelty, and serendipity.

Another challenge is the difficulty of capturing real-world user behavior. Offline evaluations are typically less expensive and time-consuming than online evaluations, but they may not accurately reflect real-world user behavior. Online evaluations provide more realistic feedback, but they can be more expensive and time-consuming to conduct. There is a need for more sophisticated evaluation methods that can accurately capture real-world user behavior. Finally, evaluating the impact of sustainability and fairness in tourism recommender systems requires new methodologies that consider the social and environmental consequences of recommendations. This is a complex and challenging task, as it requires considering the perspectives of multiple stakeholders, including tourists, local communities, and the environment.

7. Challenges and Limitations of Tourism Recommender Systems

a. Cold Start Problem

The cold-start problem is a significant challenge in tourism recommender systems, occurring when the system has insufficient information about new users or items to provide accurate recommendations [13], [14]. This can be particularly problematic in the tourism domain, where new destinations, attractions, and activities are constantly emerging. Demographic information can be used to address the cold-start problem for new tourists, providing initial recommendations based on their demographic profile [14]. Hybrid systems can also mitigate the cold start issue by combining different approaches, leveraging the strengths of both collaborative filtering and content-based filtering [34]. Additionally, starting from simple content-based recommendations and iteratively adding other methods can help address the cold-start problem, gradually building up user profiles and item information over time [15].

The cold-start problem can be particularly challenging for collaborative filtering algorithms, which rely on historical data to make recommendations. When a new user joins the system, there is no historical data available to predict their preferences. Similarly, when a new item is added to the system, there are no ratings or reviews available to assess its quality. This can lead to poor recommendations and a negative user experience. Content-based filtering algorithms can help to mitigate the cold-start problem by relying on the attributes of items, rather than the ratings of other users. However, content-based filtering algorithms can also suffer from the cold-start problem if there is insufficient information available about the attributes of new items.

Hybrid systems can address the cold-start problem by combining the strengths of both collaborative filtering and content-based filtering. For example, a hybrid system might use content-based filtering to provide initial recommendations for new users and items, and then switch to collaborative filtering as more data becomes available. Demographic information can also be used to address the cold-start problem by providing initial recommendations based on the user's demographic profile. This can be particularly useful in the tourism domain, where demographic factors such as age, gender, and income can be strong predictors of user preferences.

b. Data Sparsity

Data sparsity is another common challenge in tourism recommender systems, occurring when there are few ratings or interactions available for users and items [13]. This can result in unreliable recommendations and reduced system performance. Sparsity can be addressed by improving traditional approaches and blending them with different types of information, such as contextual data and social network information [13]. Hybrid recommendation methods also solve challenges like sparsity by combining multiple techniques [34]. Furthermore, modeling user profiles and recommended items can help mitigate the effects of sparsity, providing more accurate and personalized recommendations even with limited data [13].

Data sparsity can be particularly challenging for collaborative filtering algorithms, which rely on a dense matrix of user-item interactions to make recommendations. When the matrix is sparse, there are few users who have rated the same items, making it difficult to identify users with similar preferences. This can lead to poor recommendations and reduced system performance. Content-based filtering algorithms are less susceptible to data sparsity, as they rely on the attributes of items, rather than the ratings# Recommender Systems in Tourism: A Systematic Review

1. Introduction to Recommender Systems in Tourism

a. Overview of Recommender Systems

Recommender systems are designed to filter through large volumes of information and provide personalized suggestions to users, helping them find items or content that align with their preferences and needs [1]. These systems have become essential tools for enhancing user experiences across various domains, including e-commerce, entertainment, and tourism [2]. By analyzing user data and item characteristics, recommender systems aim to reduce information overload and improve decision-making processes. In the context of tourism, these systems play a crucial role in helping users navigate the vast amount of available data to plan their trips effectively [3], [4]. Recommender systems reduce information overload and provide travel recommendations to tourists, making it easier for them to discover destinations, accommodations, and activities that match their interests [5].

b. Importance of Tourism Recommender Systems

The tourism industry has significant potential for economic growth, and recommender systems can play a vital role in improving the quality of services by providing tailored recommendations for specific attractions [6]. Tourism recommender systems (TRSs) address the challenge of information overload by helping users find interesting destinations and plan their trips more efficiently [7], [8]. These systems use various techniques to personalize the travel planning process, making it easier for tourists to discover and select options that align with their preferences. Machine learning algorithms are increasingly used in recommender systems to design and implement personalized tourism experiences [9]. Furthermore, recommender systems in tourism can promote sustainable practices by suggesting eco-friendly activities and less popular destinations, contributing to a more balanced and responsible tourism industry [10].

c. Scope and Objectives of the Review

This systematic review aims to evaluate the most relevant and accurate techniques used in tourism recommender systems, providing insights into their strengths and limitations [6]. The review identifies trends in tourism recommender systems, with a focus on machine learning and deep learning techniques that are used to enhance personalization and accuracy [11]. The primary objective is to provide a comprehensive overview of the recommendation approaches employed in the tourism domain and to propose architectures for hybrid systems that can combine the benefits of multiple techniques [3], [4]. This systematic review summarizes the state-of-the-art methods and techniques developed by scientists in the field of tourism recommender systems, offering a framework for future research and development [12].

2. Types of Recommender Systems Used in Tourism

a. Collaborative Filtering

Collaborative filtering is a widely used technique in recommender systems that recommends items based on the preferences of users with similar tastes [13], [14]. This approach leverages the collective intelligence of a user base to identify items that a particular user might find interesting. User-based collaborative filtering, enhanced with demographic information, can provide precise recommendations for touristic sites by considering the characteristics and preferences of different user groups [14]. This method is particularly effective because it tailors recommendations to both new and registered tourists, using historical ratings and demographic data to generate relevant suggestions [14]. Geospatial collaborative recommender systems further refine this approach by incorporating spatial information to identify interesting travel destinations, taking into account the geographical context of user preferences and travel patterns [7].

b. Content-Based Filtering

Content-based filtering is another key technique used in recommender systems, where items are recommended based on the user's preferences and the features of the items themselves [13]. This approach analyzes user profiles and item descriptions to identify matches between user interests and item characteristics. Content-based recommendations are often used as a starting point in tourism recommender systems, providing a baseline set of suggestions that can be further refined using other techniques [15]. Content-based filtering analyzes user data to generate recommendations for destinations, attractions, accommodations, and activities, ensuring that the suggestions align with the user's specific interests and needs [9]. Content-based systems are particularly useful when combined with demographic data, allowing for a more nuanced understanding of user preferences and a more personalized recommendation experience [16].

c. Hybrid Approaches

Hybrid recommender systems combine multiple recommendation techniques to enhance performance and overcome the limitations of individual methods [6], [17]. These systems can fuse two or more techniques to leverage their respective strengths, resulting in more accurate and robust recommendations [18]. A hybrid multi-criteria approach, for example, can provide destination recommendations that take into account various factors such as user preferences, item characteristics, and contextual information [19]. Hybrid systems often integrate collaborative filtering and content-based filtering to improve recommendation accuracy, creating a more comprehensive and personalized recommendation experience [9].

3. Context-Aware Recommender Systems in Tourism

a. Definition and Importance of Context-Awareness

Context-aware recommender systems enhance the relevance and accuracy of recommendations by considering the user's current context, such as their location, time, and social environment [20]. Incorporating contextual information into the recommendation process allows systems to better understand the user's needs and preferences at a given moment, resulting in more personalized and useful suggestions [21]. Context-aware systems are increasingly becoming the standard in state-of-the-art recommender systems, as they provide a more dynamic and adaptive approach to personalization [15]. Context-aware recommender systems improve traditional recommender systems by using the context of information, leading to more effective and user-centric recommendations [22].

b. Types of Contextual Information

Contextual information encompasses a wide range of factors, including user demographics, item attributes, and spatial data, which can influence the relevance of recommendations [2], [18]. Location is a key contextual factor, particularly for mobile users who are accessing information while on the move, making it essential to consider the user's current geographical context [23]. Temporal context, such as the time of day or season, can also significantly influence recommendations, as user preferences and needs may vary depending on the time of year or day [16]. Furthermore, emotional states can be used to improve recommendations, allowing systems to tailor suggestions to the user's current mood and emotional condition [24].

c. Techniques for Incorporating Context

Various techniques are used to incorporate contextual information into recommender systems, including ontology-based systems, spreading activation techniques, deep learning, and the use of auxiliary information. Ontology-based systems use ontologies to group items and facilitate context-aware filtering, providing a structured and semantic approach to understanding the relationships between items and contextual factors [15], [8]. Spreading activation techniques can contextualize user preferences and learn profiles dynamically, allowing the system to adapt to changing user needs and preferences in real-time [8]. Deep learning and embeddings are used to improve context-aware recommender systems, enabling the system to learn complex patterns and relationships from large datasets [22]. Auxiliary information, including user demographics and item attributes, can further enhance recommender systems by providing additional context and insights into user preferences [2].

4. The Role of Machine Learning in Tourism Recommender Systems

a. Machine Learning Algorithms

Machine learning algorithms are fundamental to the operation of modern recommender systems, enabling the generation of personalized recommendations based on user data and item characteristics [9], [17]. Algorithms such as collaborative filtering and content-based filtering are used to analyze user preferences and item features, identifying patterns and relationships that can be used to predict user interests. These algorithms are analyzed to solve problems and improve the quality of recommendations for massive datasets, ensuring that the system can handle the scale and complexity of real-world tourism data [25]. Machine learning techniques are used to evaluate the relevance score of each item-user pair, providing a quantitative measure of the match between user preferences and item characteristics [26]. Machine learning algorithms can be trained on large datasets and continuously improve their effectiveness by incorporating new data and user feedback, ensuring that the system remains up-to-date and responsive to changing user needs [9].

b. Deep Learning Applications

Deep learning architectures are increasingly used in recommender systems to generate recommendations and improve context-aware systems, leveraging the ability of deep neural networks to learn complex patterns and relationships from data [22]. Deep learning techniques have become a research focus in recommender systems, driven by their potential to enhance personalization and accuracy [27]. Deep neural networks (DNNs) have achieved significant advancements in enhancing recommender systems, offering improved performance compared to traditional machine learning algorithms [28]. Deep learning models can incorporate textual side information about users and items, allowing the system to understand the semantic content of user reviews and item descriptions, leading to more informed and accurate recommendations [28].

c. Specific Algorithms and Techniques

Specific algorithms and techniques, such as Adaptive Neuro-Fuzzy Inference System (ANFIS) and Particle Swarm Optimization (PSO), can be used to optimize multi-criteria recommender systems, enhancing their ability to handle complex decision-making processes [29]. Embeddings are used to represent data in recommender systems and improve performance, allowing the system to capture the semantic relationships between users and items in a low-dimensional space [22]. Case-based reasoning is used to provide recommendations based on previously solved cases, leveraging the knowledge and experience gained from past interactions to guide future recommendations [30]. Opinion mining and contextual information extraction techniques are used to incorporate user reviews into recommender systems, allowing the system to understand the sentiment and context of user feedback, leading to more nuanced and accurate recommendations [20].

5. Data Sources and Collection Methods

a. Types of Data Used

Recommender systems rely on various types of data to generate personalized recommendations, including user data, item data, and contextual data. User data, including preferences, behavior, demographic profiles, and social network judgments, is used to develop recommender systems that align with individual user needs and interests [9], [2]. Item data, including attributes and features, is used in content-based filtering to match user preferences with item characteristics, ensuring that recommendations are relevant and informative [13]. Contextual data, such as location and time, is used in context-aware systems to adapt recommendations to the user's current situation, providing a more dynamic and personalized experience [20]. Tourist data is generated across various sectors, including hotels, restaurants, transportation, and heritage sites, providing a rich and diverse source of information for building effective recommender systems [3].

b. Data Collection Techniques

Data collection techniques for recommender systems vary depending on the type of data being collected and the sources available. Data is often collected from online platforms, including social media and e-commerce sites, where users generate vast amounts of data through their interactions and activities [20], [6]. Websites and online travel agencies (OTAs) are significant sources of tourist data, providing information on destinations, accommodations, and activities [3]. Sensors and wearables can collect data for emotion recognition in tourism recommender systems, allowing the system to understand the user's emotional state and tailor recommendations accordingly [24]. User reviews and ratings are extracted from platforms like TripAdvisor, providing valuable insights into user experiences and preferences [14], [29].

c. Data Preprocessing and Integration

Data preprocessing and integration are essential steps in the development of recommender systems, ensuring that the data is clean, consistent, and ready for analysis. Data preprocessing involves cleaning, transforming, and integrating data from various sources, addressing issues such as missing values, inconsistencies, and duplicates [15]. Natural language processing (NLP) is used to classify items and associate them with user preferences, allowing the system to understand the semantic content of user reviews and item descriptions [15]. Ontologies are used to structure and organize data, facilitating the association between user preferences and items, and providing a semantic framework for understanding the relationships between different entities [15]. Embeddings are used to represent data in a format suitable for machine learning algorithms, allowing the system to learn complex patterns and relationships from data [22].

6. Evaluation Metrics and Methods

a. Common Evaluation Metrics

Evaluating the performance of recommender systems requires the use of appropriate metrics that can quantify the accuracy, relevance, and effectiveness of the recommendations. Precision, recall, and F1-score are commonly used to evaluate the quality of recommendations, measuring the proportion of relevant items that are correctly recommended, the proportion of all relevant items that are recommended, and the harmonic mean of precision and recall [14], [19]. Root Mean Squared Error (RMSE) is used to evaluate prediction accuracy, measuring the difference between the predicted ratings and the actual ratings provided by users [14], [19]. Normalized Discounted Cumulative Gain (NDCG) is used to evaluate the ranking quality of recommendations, measuring the relevance of items at different positions in the recommendation list, with higher weight given to items at the top [31]. Beyond-accuracy qualities are rarely assessed, highlighting the need for more diverse evaluation metrics that consider factors such as fairness, diversity, and novelty [31].

b. Offline vs. Online Evaluation

Evaluation of recommender systems can be conducted using offline or online methods, each with its own advantages and limitations. Offline experimentation is the predominant experiment type in research on recommender systems, allowing researchers to evaluate different algorithms and techniques using historical data [31]. Online evaluations are primarily used in combination with other experimentation methods, providing real-world feedback on the performance of the system and allowing for A/B testing of different approaches [31]. User studies are conducted to evaluate different aspects of recommender systems from the users' perspective, providing qualitative insights into user satisfaction and preferences [32]. Acceptance testing is used to measure the easiness of use, efficiency, and correctness of recommendations, ensuring that the system meets the needs and expectations of its users [30].

c. Challenges in Evaluation

Evaluating recommender systems presents several challenges, including the limited availability of widely used datasets, the need for standardization of non-standard evaluation measures, and the limitations of current evaluation approaches. Few datasets are widely used, and many datasets are used in only a few papers each, making it difficult to compare the performance of different systems across different studies [31]. There is a need for standardization of non-standard evaluation measures, ensuring that the results of different studies are comparable and that the evaluation process is rigorous and consistent [13]. Current evaluation approaches have significant limitations and must be addressed in future studies, including the need to consider factors such as fairness, diversity, and novelty [33]. Evaluating the impact of sustainability and fairness in tourism recommender systems requires new methodologies that can capture the complex interactions between users, items, and the broader social and environmental context [10].

7. Challenges and Limitations of Tourism Recommender Systems

a. Cold Start Problem

The cold-start problem is a significant challenge in recommender systems, occurring when the system has insufficient information about new users or items to make accurate recommendations [13], [14]. This issue is particularly relevant in tourism, where new destinations, attractions, and activities are constantly emerging. Demographic information can be used to address the cold-start problem for new tourists, allowing the system to make initial recommendations based on their demographic characteristics and preferences [14]. Hybrid systems can mitigate the cold start issue by combining different approaches, such as content-based filtering and collaborative filtering, to leverage the strengths of each method [34]. Starting from simple content-based recommendations and iteratively adding other methods can help address the cold-start problem, allowing the system to gradually learn about user preferences and item characteristics over time [15].

b. Data Sparsity

Data sparsity is another common challenge in recommender systems, occurring when there are few ratings or interactions available for users and items, making it difficult to identify patterns and relationships [13]. Sparsity can be addressed by improving traditional approaches and blending them with different types of information, such as contextual data and social network information, to enrich the data available to the system [13]. Hybrid recommendation methods solve the challenges like sparsity by combining different techniques and leveraging their respective strengths [34]. Modeling user profiles and recommended items can help mitigate the effects of sparsity, allowing the system to make more accurate recommendations even when data is limited [13].

c. Scalability and Real-time Recommendations

Scalability is a significant challenge when dealing with large datasets and a high volume of website visits, requiring efficient algorithms and data processing techniques to ensure that the system can handle the load [13], [25]. Real-time recommendations require efficient algorithms and data processing techniques to ensure that the system can respond quickly to changing user needs and preferences [32]. Dynamic adaptation of user preferences and results is important for real-time recommendations, allowing the system to adjust its suggestions based on the user's current context and behavior [32]. Big data technologies and artificial intelligence can be used to develop scalable tourism recommender systems that can handle the volume and complexity of real-world tourism data [4].

8. Fairness and Ethical Considerations

a. Individual Fairness

Fairness in recommender systems is an increasingly important consideration, ensuring that similar users receive similar recommendations and that the system does not discriminate against certain groups or individuals [35]. Individual fairness focuses on providing equitable treatment to all users, regardless of their demographic characteristics or other sensitive attributes [35]. Fairness considerations are essential in tourism recommender systems to avoid discriminatory outcomes, such as recommending less desirable destinations or accommodations to certain groups of users [35]. Addressing unique challenges and ensuring fairness within the tourism domain is crucial, requiring careful attention to the design and evaluation of recommender systems [35].

b. Multi-Stakeholder Fairness

Multi-stakeholder fairness considers the perspectives and needs of different stakeholders, including end-users, item providers, and platforms, ensuring that the system benefits all parties involved [35]. Achieving goals for one stakeholder may conflict with those of another, resulting in trade-offs that must be carefully considered and balanced [35]. A multi-faceted concept requires consideration of perspectives and needs of different stakeholders to ensure fair outcomes for them, promoting a more equitable and sustainable tourism ecosystem [35]. Strategies for mitigating unfairness and examining the applicability of solutions from other domains are needed to address the complex challenges of multi-stakeholder fairness [35].

c. Bias and Discrimination

Data bias can lead to unfair or discriminatory recommendations, perpetuating existing inequalities and biases in the tourism industry [36]. Addressing bias in recommender systems is essential to ensure equitable outcomes, requiring careful attention to the data used to train the system and the algorithms used to generate recommendations [36]. Psychological factors such as personality, emotions, and decision biases can affect the outcome of the recommendation process, highlighting the need to consider these factors in the design of recommender systems [37]. Integrating psychological aspects into systems helps better predict users' item preferences and increase satisfaction, leading to more personalized and effective recommendations [37].

9. Emerging Trends and Future Directions

a. Integration of Large Language Models (LLMs)

Large Language Models (LLMs) have revolutionized the fields of NLP and AI, offering new opportunities to enhance recommender systems by leveraging their ability to understand and generate human-like text [28]. LLMs can improve the ability to capture textual side information about users and items, allowing the system to understand the semantic content of user reviews, item descriptions, and other textual data [28]. LLMs can be used for pre-training, fine-tuning, and prompting in recommender systems, enabling the system to learn from large amounts of text data and adapt to different recommendation tasks [28]. Future research directions include leveraging LLMs as feature encoders and exploring advanced techniques for enhancing recommender systems, such as using LLMs to generate personalized explanations for recommendations [28].

b. Sustainability-Oriented Recommender Systems

Sustainability-oriented recommender systems integrate sustainability-driven goals into the design to improve tourism development, promoting responsible and eco-friendly tourism practices [26]. These systems aim to reduce the tensions created by interactions between tourists and the environment, mitigating the negative impacts of tourism on natural and cultural resources [26]. Sustainable tourism practices are becoming increasingly important, and recommender systems can play a crucial role in promoting sustainability by suggesting sustainable activities and less popular areas [10]. Incorporating sustainable options in recommendations can encourage tourists to visit sustainable and less popular areas, helping to address issues such as overtourism and undertourism in the travel and tourism industry [10].

c. Smart Tourism and the Internet of Things (IoT)

The Internet of Things (IoT) transforms smart tourism by boosting operational efficiency and enriching traveler experiences, creating a more connected and personalized tourism ecosystem [38], [39]. IoT technologies, including big data, sensors, cloud computing, and AI, establish the foundation for an IoT-enabled tourism ecosystem, enabling the collection and analysis of vast amounts of data from various sources [38]. IoT applications in tourism include recommender systems, smart cities, and electronic ticketing, providing a seamless and integrated experience for tourists [38]. Future directions include enhancing security, exploring blockchain integration, and improving interactions between tourists and systems, ensuring that the benefits of IoT are realized in a responsible and sustainable manner [38].

10. Conclusion

a. Summary of Key Findings

Recommender systems are essential tools for personalized user experiences in the tourism domain, helping tourists navigate the vast amount of available information and plan their trips more effectively [2], [9]. Various recommendation approaches, including collaborative filtering, content-based filtering, and hybrid methods, are used in tourism, each with its own strengths and limitations [13], [6]. Context-aware recommender systems enhance prediction tasks by incorporating contextual information, such as location, time, and social environment, leading to more relevant and personalized recommendations [20], [21]. Machine learning and deep learning techniques play a crucial role in improving the accuracy and personalization of recommendations, enabling the system to learn from large amounts of data and adapt to changing user needs [9], [27].

b. Implications for the Tourism Industry

Recommender systems can help tourists navigate the vast amount of available data and plan trips more effectively, reducing information overload and improving decision-making [3], [4]. Personalized recommendations can improve user experience and increase revenue for tourism businesses, leading to greater customer satisfaction and loyalty [9]. Sustainability-oriented recommender systems can promote sustainable tourism practices and reduce the negative impacts of tourism, contributing to a more responsible and eco-friendly tourism industry [10], [26]. The integration of IoT technologies can further enhance the efficiency and personalization of tourism services, creating a more seamless and integrated experience for tourists [38], [39].

c. Future Research Directions

Future research should focus on addressing the challenges of cold start, data sparsity, and scalability, developing new techniques and algorithms to overcome these limitations [13], [25]. Further exploration of fairness and ethical considerations in tourism recommender systems is needed to ensure that the system does not discriminate against certain groups or individuals and that it benefits all stakeholders involved [35]. The integration of large language models (LLMs) and other emerging technologies should be explored to enhance recommender systems, leveraging their ability to understand and generate human-like text [28]. Developing comprehensive construction methodologies and evaluation across multiple quality dimensions for tourism knowledge graphs (TKGs) should be prioritized to improve the organization and leveraging of machine-readable knowledge in the tourism domain [40].