Clip recommendation based on topic similarity

Abstract

We propose a clip recommendation technology based on topic similarity. Topics of a clip can represent semantics of each contents. When the topic distributions for clips are similar, it means they are alike. In this paper, we propose a system to learn topic distributions for broadcast contents and link clips based on topics similarity of each clip. The higher the similarity is among the clips, the higher the semantic is among them. This system can be adopted clip recommendation with audiences viewing history and their interest.