Scene emotion detection using closed caption based on hierarchical attention network

This paper proposes a method of emotion detection using closed caption. Closed caption is a one of the texts related with video contents, which can be obtained actual dialogue information. Emotion information in video contents has been used in various service such as video recommendation, video retrieval. To detect scene emotion, we apply hierarchical learning structure of closed caption, such as word-sentence-document. In this paper, we introduce scene emotion detection method based on bidirectional LSTM and hierarchical attention network model. In experiments, we show the results of emotion detection using closed caption which from Korean movie.