Recommendation and Planning Systems for Tourism Products and Services
Recommender Systems (RSs) are information search and filtering tools that provide suggestions for items to be of use to a user. They have become common in a large number of Internet applications, helping users to make better choices while searching for news, music, vacations or financial investments. RSs exploit data mining and information retrieval techniques to predict to what extent an item suits the user needs and wants and recommend those items with the largest predicted fit score. Tourism applications have focussed on recommending: destinations, activities, accommodations and routes.
Motivation and Diffusion
The explosive growth and variety of information available on the Web and the rapid introduction of new e-business and social services (buying products, product comparison, auction, forums, social networking, multimedia fruition) has created such a richness of choices, that instead of producing a benefit, this overabundance risks to backfire. If dozens of different types of jams are likely to confuse and paralyze a buyer, as it is illustrated in (Schwartz, 2004), hundreds of hotels or holiday destinations are simply impossible to evaluate and compare if the ultimate goal is just to select the most appropriate one (Ricci 2011). Such a scenario motivated the introduction of Recommender Systems (RSs) (Ricci et al, 2011; Konstan and Riedl, 2012). Recommender systems play an important role in highly rated Internet sites, such as: Amazon.com, YouTube, Netflix, Yahoo, Tripadvisor, Last.fm, and IMDb.
Definition and Function
RSs are information search and filtering tools that provide suggestions for items to be of use to a user. They help users to make better choices while searching for news, music, vacations, or financial investments. 'Item' is the general term used to denote what the system recommends to its users, and a specific RS normally focuses on one type of items (e.g., movies, or vacations). Accordingly, its core algorithmic component and its graphical user interface are customized to provide useful and effective suggestions for that specific type of items.
In their simplest form, personalized recommendations are offered as customized lists of items. In performing this selection the system tries to predict what the most suitable products or services (items) are, based on the user's characteristics and preferences. In order to complete such a computational task, a RS must elicit from users such characteristics and preferences, either along the full history of previous interactions with the users, or exploiting information entered by the users at the time the recommendation is requested. Moreover, such information can be either explicitly expressed, e.g., as ratings for products, or can be inferred by interpreting user actions. For instance, the navigation to a particular page can be interpreted as an implicit sign of preference for the items shown on that page.
Study and Research on Recommender Systems
The study of recommender systems is relatively new compared to research in other classical information system tools and techniques (e.g., databases or search engines). Recommender systems emerged as an independent research area in the mid 1990s (Goldberg et al, 1992; Resnick et al, 1994), and it is still fast growing. Research works on RSs are published in major conferences on machine learning (ICML, KDD, NIPS), information retrieval (SIGIR, WISDM, CIKM), intelligent user interfaces (IUI), personalization (UMAP), and Tourism (ENTER). A specific ACM conference on Recommender Systems has been launched on 2007, and every year it attracts more and more submissions and attendees. In total, thousands of papers are published every year on this subject.
A general computational model for recommender systems is described in (Adomavicius and Tuzhilin, 2005). A RS is defined as a machinery implementing a real valued function defined on the product space of the users and items r* :U×I→R that predicts how a pair consisting of a user u∈U and an item i∈I is mapped to the evaluation r∗(u,i) of the user u for the item i. Then, having predicted evaluations of users for items, a RS recommends to a user u the items i with the largest predicted evaluations r∗(u,i). Evaluations are called ratings of users for items in Collaborative Filtering RSs (see next section on Prediction Techniques). A RS computes the prediction r∗(u,i) on the base of a collection of observations: these are interactions between users and items, and they provide the system with information about the users’ preferences. In many cases these interactions produce explicit evaluations performed by some users on some items. In some other cases, more complex types of relationships are observed, for instance the relative preference of a user for an item when it is compared to another item.
In order to implement the item evaluation prediction function RSs can exploit a range of techniques. This has been the major topic of research in RSs (Ricci et al, 2011). Recommendation techniques vary in terms of the addressed domain, the knowledge used, and the recommendation algorithm, i.e., essentially how the item evaluation prediction is actually computed. The following taxonomy was introduced by (Burke, 2007):
Content-based: The system implements for each user a 'classifier' that learns to evaluate (classify) higher the items that are similar to the ones that the user evaluated higher in the past. The similarity of items, or more in general the item classification rule, is calculated based on the features associated with the compared items. For example, if a user has systematically positively rated places of interest that belong to the "museum" type, then the system can learn that other museums should have a high value for that user (Lops et al, 2011).
Collaborative filtering: The simplest and original implementation of this approach predicts that the active user, i.e., the user asking for recommendations, will evaluate higher the items that other users with similar tastes liked in the past (Desrosiers and Karypis, 2011). The similarity in taste of two users is calculated based on the similarity of the evaluations' history of the users. Collaborative filtering is probably the most popular and widely implemented technique in RSs. The latest approaches to CF use latent factor models, such as matrix factorization (e.g., using Singular Value Decomposition, SVD). These methods map both items and users to the same latent factor space. Then the predicted evaluation of a user for an item is basically computed by the dot multiplication of their representative vectors, which gives a kind of similarity between the user and the item in this common representation space (Koren and Bell, 2011).
Demographic: These techniques predict item evaluations based on the demographic profile of the user. The assumption is that different recommendations should be generated for different demographic niches. Many tourism Web sites adopt simple and effective personalization solutions based on demographics. For example, users are dispatched to particular Web sites based on their language or country. Or suggestions may be customized according to the age of the user.
Knowledge-based: Knowledge-based systems predict item evaluations based on specific domain knowledge about how certain item features meet user's needs and preferences, and ultimately how the item is useful for the user. Notable knowledge-based recommender systems are case-based (Bridge et al, 2006). In these systems a similarity function estimates how much the user needs (problem description) match the recommendations (solutions of the problem). Here the similarity score can be directly interpreted as the predicted item evaluation of the user. Another group of knowledge-based systems uses constraints, to represent user preferences and to find relevant items (Jannach et al, 2010).
Community-based: In this type of systems item evaluation predictions are based on the preferences of the user's friends. Evidence suggests that people tend to rely more on recommendations from their friends than on recommendations from similar but anonymous individuals. This observation, combined with the growing popularity of open social networks, is generating a rising interest in community-based systems or social recommender systems (Golbeck, 2006). This type of RSs techniques acquires and exploits information about the social relations of the users and the preferences of the user's friends. The item evaluation predictions are based on ratings that were provided by the userâ€™s friends.
Hybrid recommender systems: These RSs are based on the combination of the above mentioned techniques. A hybrid system combining techniques A and B tries to use the advantages of A to fix the disadvantages of B. For instance, CF methods suffer from new-item problems, i.e., they cannot generate evaluation predictions for items that have no ratings. This does not limit content-based approaches since the prediction for new items is based on their description (features) that are typically easily available. Given two (or more) basic RSs techniques, several ways have been proposed for combining them to create a new hybrid system (Burke, 2007).
The research on RSs is still very active and numerous issues and challenges are still open. We want to list here some of them, with the obvious caveat that this list cannot be complete and is influenced by our personal knowledge and vision. The reader is also referred to (Konstan and Riedl, 2012) for another discussion on the future of recommender systems.
- Group Recommenders: deal with situations when it would be good if the system could recommend information or items that are relevant to a group of users rather than to an individual (Masthoff, 2011).
- Proactive recommender systems: can decide to push recommendations even if not explicitly requested. In this scenario, the system therefore must predict what to recommend, but also when and how to 'push' its recommendations.
- Active learning: RSs need to actively look for new data during the operational life (Rubens et al, 2011). This issue was normally neglected on the assumption that there is not much space for controlling what data (e.g., ratings) the system can collect, because these decisions are autonomously taken by the users when visiting the system.
- Privacy preserving: RSs exploit user data to generate personalized recommendations. This can clearly have a negative impact on the privacy of the users and the users may start feeling that the system knows too much about their true preferences (Kobsa, 2008).
- Diversity: assuring the diversity of the items recommended to a target user is an important feature of a recommender system (Vargas and Castells, 2011). For instance, it is more likely that the user will find a suitable item in a recommendation list, if there is a certain degree of diversity among the recommendations.
- Generic user models: generic user models (Kobsa, 2007), and cross domain recommender systems are able to mediate user data (item evaluations) through different systems and application domains (Berkovsky et al, 2008) hence can be bootstrapped in a domain using data coming from a different domain.
- Sequential Recommendations: recommenders may optimize a sequence of recommendations (Baccigalupo and Plaza, 2006) or a sequence of actions that brings to a recommendation (Mahmood et al., 2009). For instance a sequence of week-end vacations suggested to a tourist.
Adomavicius G, Tuzhilin A (2005) Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions. IEEE Transactions on Knowledge and Data Engineering 17(6):734-749
Baccigalupo C, Plaza E (2006) Case-based sequential ordering of songs for playlist recommendation. In: Roth-Berghofer T, G Ìˆoker MH, Gu Ìˆvenir HA (eds) ECCBR, Springer, Lecture Notes in Computer Science, vol 4106, pp 286-300
Berkovsky S, Kuflik T, Ricci F (2008) Mediation of user models for enhanced personalization in recommender systems. User Modeling and User-Adapted Interaction 18(3):245-286
Bridge D, G Ìˆoker M, McGinty L, Smyth B (2006) Case-based recommender systems. The Knowledge Engineering review 20(3):315-320
Burke R (2007) Hybrid web recommender systems. In: The Adaptive Web, Springer Berlin / Heidelberg, pp 377-408
Desrosiers C, Karypis G (2011) A comprehensive survey of neighborhood-based recommendation methods. In: Ricci F, Rokach L, Shapira B, Kantor PB (eds) Recommender Systems Handbook, Springer, pp 107-144
Golbeck J (2006) Generating predictive movie recommendations from trust in social networks. In: Trust Management, 4th International Conference, iTrust 2006, Pisa, Italy, May 16-19, 2006, Proceedings, pp 93-104
Goldberg D, Nichols D, Oki BM, Terry D (1992) Using collaborative filtering to weave an information tapestry. Commun ACM 35(12):61-70
Jannach D, Zanker M, Felfernig A, Friedrich G (2010) Recommender Systems: An Introduction. Cambridge University Press
Kobsa A (2007) Generic user modeling systems. In: Brusilovsky P, Kobsa A, Nejdl W (eds) The Adaptive Web, Springer, Lecture Notes in Computer Science, vol 4321, pp 136-154
Kobsa A (2008) Privacy-enhanced personalization. In: Proceedings of the Twenty-First International Florida Artificial Intelligence Research Society Conference, May 15-17, 2008, Coconut Grove, Florida, USA, AAAI Press, p 10
Konstan JA, Riedl J (2012) Recommender systems: from algorithms to user experience. User Model User-Adapt Interact 22(1-2):101-123
Koren Y, Bell R (2011) Advances in collaborative filtering. In: Ricci F, Rokach L, Shapira B (eds) Recommender Systems Handbook, Springer Verlag, pp 145-186
Lops P, de Gemmis M, Semeraro G (2011) Content-based recommender systems: State of the art and trends. In: Ricci F, Rokach L, Shapira B, Kantor PB (eds) Recommender Systems Handbook, Springer Verlag, pp 73-105
Mahmood T, Ricci F, Venturini A (2009) Improving recommendation effectiveness by adapting the dialogue strategy in online travel planning. International Journal of Information Technology and Tourism 11(4):285-302
Masthoff J (2011) Group recommender systems: Combining individual models. In: Recommender Systems Handbook, Springer, pp 677-702
Resnick P, Iacovou N, Suchak M, Bergstrom P, Riedl J (1994) Grouplens: An open architecture for collaborative filtering of netnews. In: Proceedings ACM Conference on Computer-Supported Cooperative Work, pp 175-186 Ricci F (2011) Mobile recommender systems. International Journal of Information Technology and Tourism 12(3):205-231
Ricci F, Rokach L, Shapira B, Kantor PB (2011) Recommender Systems Handbook. Springer
Rubens N, Kaplan D, Sugiyama M (2011) Active learning in recommender systems. In: Ricci F, Rokach L, Shapira B, Kantor PB (eds) Recommender Systems Handbook, Springer Verlag, pp 735-767
Schwartz B (2004) The Paradox of Choice. ECCO, New York
Vargas S, Castells P (2011) Rank and relevance in novelty and diversity metrics for recommender systems. In: Proceedings of the 2011 ACM Conference on Recommender Systems, RecSys 2011, Chicago, IL, USA, October 23-27, 2011, ACM, pp 109-116
- Footnote: This note is based on a short extract from a longer article titled "Recommender Systems, Models and Techniques" that will be published in: Encyclopedia of Social Network Analysis and Mining. Alhajj, Reda; Rokne, Jon (Eds.), Springer Verlag, 2014.