Exploiting multimodal interactions in recommender systems with ensemble algorithms

The increasing of products, information and services based on users׳ profiles has made recommender systems to be increasingly present, easing the selection and retention of users in services on the Web. However, optimizations must be performed in such systems mainly regarding the modeling of users׳ profiles. Preferences are generally modeled superficially, due to the scarcity of data collected, as notes or comments, or the inductive information susceptible to errors. This manuscript proposes a recommender tool with three ensemble approaches based on multimodal interactions that combines different types of users׳ feedback processed individually by traditional recommendation algorithms. The approaches have been developed to improve the quality of predictions in recommender systems, considering a large number of user information.