Entry point: The item that the user chooses as a starting point can be considered a strongly positive preference, since the user is looking for something similar to it. Ending point: The final selection or buying decision can also be considered a positive rating. Tweaking: When a user critiques a returned item and moves on to something different, we can consider this a negative rating. Browsing: If the user navigates to other items in the returned set, we can consider this a weak negative rating: if the user truly liked the item he or she would probably not browse further. These heuristics are somewhat weak, since we sometimes find users who are exploring the system to see what it can do, applying tweaks not to get a specific recommendation, but to see what will come back.
Collaborative filtering (CF) is a technique for recommending items to a user's attention based on similarities between the past behavior of the user and that of other users. A canonical example is the GroupLens system that recommends news articles based on similarities between users' reading behavior (Resnick, et al. 1994). This technique has been applied to many areas from consumer products to web pages (Resnick Varian, 1997; Kautz, 1998), and has become standard marketing technique in electronic commerce. The input to a CF system is a triple consisting of a user, an object that the user has an opinion about, and a rating that captures that opinion: u, o, r(u,o) . As ratings for a given user are accumulated, it becomes possible to correlate users on the basis of similar ratings and make predictions about unrated items on the basis of historical similarity.
We describe a recommender system which uses a unique combination of content-based and collaborative methods to suggest items of interest to users, and also to learn and exploit item semantics. Recommender systems typically use techniques from collaborative filtering, in which proximity measures between users are formulated to generate recommendations, or content-based filtering, in which users are compared directly to items. Our approach uses similarity measures between users, but also directly measures the attributes of items that make them appealing to specific users. This can be used to directly make recommendations to users, but equally importantly it allows these recommendations to be justified. We introduce a method for predicting the preference of a user for a movie by estimating the user's attitude toward features with which other users have described that movie. We show that this method allows for accurate recommendations for a subpopulation of users, but not for the entire user population. We describe a hybrid approach in which a userspecific recommendation mechanism is learned and experimentally evaluated. It appears that such a recommender system can achieve significant improvements in accuracy over alternative methods, while also retaining other advantages.
While recommender systems are in widespread use, they still experience problems. Many recommender systems produce recommendations which the customers find unsatisfactory. Further, these systems often suffer from problems when there are not enough participants, or when new products enter the system. We perceive an opportunity for knowledge-based recommender systems to gain leverage on recommendation tasks by using explicit models of both the user of the system and the products being recommeded. This differs from previous systems which, when they use a user model, have used one that is inferred from the ratings given by that user (i.e., an implicit user model).
While recommender systems can greatly enhance the user experience, the entry barriers in terms of data acquisition are very high, making it hard for new service providers to compete with existing recommendation services. This paper proposes to build open recommender systems which can utilise Linked Data to mitigate the new-user, new-item and sparsity problems of collaborative recommender systems. We describe how to aggregate data about object centred sociality from different sources and how to process it for collaborative recommendation. To demonstrate the validity of our approach, we augment the data from a closed collaborative music recommender system with Linked Data, and significantly improve its precision and recall.