From Community Detection to Mentor Selection in Rating-Free Collaborative Filtering

International audienceThe number of resources or items that users can now access when navigating on the Web or using e-services, is so huge that these might feel lost due to the presence of too much information. Recommender systems are a way to cope with this profusion of data by suggesting items that fit the users' needs. One of the most popular techniques for recommender systems is the collaborative filtering approach that does not use any a priori information about the users, nor any data about the content of the items. Collaborative filtering relies on the preferences of items expressed by users. These are usually recorded under the form of ratings and the recommendation technique exploits these ratings. However, in many e-services, it is inappropriate to ask to rate items; it may indeed interrupt users' activity. In the absence of ratings, classical collaborative filtering techniques cannot be applied; especially the selection of like-minded users for a given user, also called his mentor users, cannot be performed. Fortunately, the behavior of users, such as their consultations, can be collected; this collection is transparent for users. In this paper, we focus on rating-free collaborative filtering: we present a new approach to perform collaborative filtering when no rating is available but when user consultations are known. We propose to take inspiration from local community detection algorithms to form communities of users and deduce the set of mentor users of a given user. These algorithms have the advantage of not only being less complex than community detection algorithms, but also of discovering overlapping communities. We adapt one state of the art algorithm so as to fit the characteristics of collaborative filtering. Experiments conducted on the two datasets show that the precision achieved by this community detection algorithm is higher then the baseline that does not perform any mentor selection. In addition, our model almost offsets the absence of ratings by exploiting a set of mentors reduced by 71\% and 99\% compared to the baseline