Δημιουργία συστήματος συνεργατικού φιλτραρίσματος για σύνολα δεδομένων έμμεσης ανατροφοδότησης
Collaborative filtering for implicit feedback datasets
Recommender systems, or recommendation systems, is a way to predict the behavioral patterns of a user, his preferences or dislikes, in order to provide personalized recommendations. These systems work based on prior explicit or implicit feedback. The much more extensively researched explicit feedback systems gather their knowledge directly from the users, in the form of a simple rating. For example, Netflix uses a rating scale, from one to five stars, to determine whether a user enjoyed a specific movie or not. On the other hand, implicit feedback systems work passively in the background, tracking different sorts of user behavior, such as browsing activity, watching habits or purchase history. Thus, in the case of implicit feedback systems, we do not have any direct indication of the user’s preferences and, specifically, we do not have any significant evidence on which items a user dislikes. The aim of this study is to analyze the unique properties of implicit feedback datasets, especially tailored to fit the retail sector, confronting a vast variety of consumer products, their price differences and user-item interaction sparsity.