Contextual TV Show Recommendation

Abstract

Recommender systems are a form of artificial intelligence that is used to suggest items to users of digital platforms. They use large data sets to infer models of users’ behavior and preferences in order to recommend items that the user may be interested in. Following the trend imposed by digital media companies and willing to adapt to the media consumption habits of their customers, TV broadcasters are starting to realize the potential of recommender systems to personalize the access to their online catalog. By understanding what viewers are watching and what they might like, TV broadcasters can improve the quality of their programming, increase viewership, and attract new viewers. In this work, we analyze one specific group of users that TV broadcasters must take into account when creating a recommender system: non-logged users. In this scenario the challenge is to use contextual information about the interaction in order to predict recommendations, as it is not feasible to use any kind of information about the user. We propose a method to leverage data from other type of users (logged users and identified devices) by using Graph Convolutional Networks in order to come up with a more accurate recommender system for unidentified users.