Machine Learning the TV Consumption: A Basis for a Recommendation System

With the continuous growth of channels and content available in a typical interactive TV service, viewers have become increasingly frustrated, struggling to select which programs to watch. Content recommendation systems have been pointed out as a possible tool to mitigate this problem, especially when applied to on-demand content. However, in linear content, its success has been limited, either due to the specificities of this type of content or due to the little integration with normal consumption behaviors. Despite that, recommendation algorithms have undergone a set of enhancements in order to improve their effectiveness, particularly when applied to the world of linear content. These improvements, focused on the use of the visualization context, paired with machine learning techniques, can represent a big advantage in the quality of the suggestions to be proposed to the viewer. The area of user experience (UX) evaluation, in interactive TV, has been also a subject of ongoing research, extending beyond the traditional usability evaluation, pursuing other dimensions of analysis such as identification, emotion, stimulation, and aesthetics, as well as distinct moments of evaluation. This paper presents the proposal for the development of a recommendation system, based on the viewing context, and a methodology for evaluating the way this system influences the UX of the viewers.