Hybrid algorithms for recommending new items

Despite recommender systems based on collaborative filtering typically outperform content-based systems in terms of recommendation quality, they suffer from the new item problem, i.e., they are not able to recommend items that have few or no ratings. This problem is particularly acute in TV applications, where the catalog of available items (e.g., TV programs) is very dynamic. On the contrary, content-based recommender systems are able to recommend both old and new items but the general quality of the recommendations in terms of relevance to the users is low. In this article we present two different approaches for building hybrid collaborative+content recommender systems, whose purpose is to produce relevant recommendations, while overcoming the new item issue. The approaches have been tested on two datasets: a version of the well--known Movielens dataset enriched with content meta--data, and an implicit dataset collected from 15'000 IPTV users over a period of six months.