Evaluating the Impact of Demographic Data on a Hybrid Recommender Model

One of the major challenges in Recommender Systems is how to predict users’ preferences in a group context. There are situations in which a user could be recommended an item appropriated for one of their groups, but the same item may not be suitable when interacting with another group. There are situations in which a user could be recommended an item appropriated for one of their groups (e.g. poetry friends), but the same item may not be suitable when interacting with another group (e.g. soccer team). We note that recommender systems should try to satisfy the group’s demands, but it should also respect the user’s individuality. The demographic data are an effective way to consider users’ characteristics, enabling analysis about group of users and their contextual constraints. In our past work we have proposed a multifaceted hybrid recommender model, which integrates a set of different user’s inputs into a unified and generic latent factor model, achieving better results than the other state-of-the-art approaches. The recommender exploits users’ demographics, such as age, gender and occupation, along with implicit feedback and items’ metadata. Depending on the personal information from users, the recommender selects content whose subject is semantically related to their interests. In this paper we evaluate our model, aiming to analyze the impact of demographic data on hybrid recommenders, in order to understand how these systems are beavering in terms of the recommendation accuracy, precision and recall, computational cost and improvements of technical results.