Uma arquitetura multimodal para recomendação baseada em conteúdo para TV digital.

Abstract

Digital TV content providers are becoming widespread, with hundreds of programs available each day. The information overload makes difficult for the user to find programs of interest. To help the user, recommender systems (RSs) are a popular path. However, applying RSs to some environments is not an easy task, either due to the lack of data or because the data available is insufficient to create accurate recommendations using standard RS approaches. In the Digital TV domain, the main information available to make recommendations is the Electronic Program Guide (EPG). The information available on EPG is limited, containing only reduced textual data, making difficult to get an accurate recommendation using standard techniques. To solve this problem, in this work we introduce an architecture that uses multimodal approach to recommend Digital TV programs, combining EPG text and visual information. We performed an experiment and demonstrated that using multimodal features the accuracy of the recommendation can be improved when compared with a recommender standard approach.