Multi-view topic model learning to generate audience metadata automatically

Abstract

In this paper, we propose a study on multi-view topic model learning to generate automatically audience metadata for clips. We use closed caption of broadcasting contents, user's subscription information and viewing history. An existing topic model has limits to being utilized for user targeted services by learning topics based on subtitles or scripts without user data. To overcome this limitation, this paper proposes a multi-view topic model learning technique using multi domain data such as closed caption of broadcast contents and viewing rating of audience groups.