RESUMO

Recommender systems have become increasingly popular and widely adopted by many sites and services. They are important tools in assisting users to filter what is relevant for them in this complex information world. There are a number of ways to build recommender systems such as content-based filtering, which recommends multimedia content to the user based on a profile containing information regarding the content, such as genre, keywords, subject, etc. These metadata are weighted according to past ratings, in order to characterize the user’s main interests. However, this approach has problems such as over-specialization and limited performance due to metadata scarcity or quality. An alternative to this problem is the collaborative filtering approach, which is based on clusters of similar users or items. The drawback is the computational effort spent to calculate similarity between users and/or items in a vectorial space composed of user ratings in a user-item matrix. Alternatively, hybrid recommenders aim at grouping the benefits of content based and collaborative filtering approaches. The downside of hybrid recommenders which primarily exploit latent factor models are i) do not consider the metadata associated to the content, which could provide significant and meaningful information about the user’s interests, and ii) usually process only one item attribute missing the exploitation of combination of the metadata available