Exploring the Ratings Prediction Task in a Group Recommender System that Automatically Detects Groups

Recommender systems produce content for users, by suggesting items that users might like. Predicting the ratings is a key task in a recommender system. This is especially true in a system that works with groups, because ratings might be predicted for each user or for the groups. The approach chosen to predict the ratings changes the architecture of the system and what information is used to build the predictions. This paper studies approaches to predict the ratings in a group recommendation scenario that automatically detects groups. Experimental results confirm that the approach to predict the ratings strongly influences the performances of a system and show that building predictions for each user, with respect to building predictions for a group, leads to great improvements in the quality of the recommendations.