Poster: Sentiment Analysis of BGM Toward Automatic BGM Selection Based on Emotion

Abstract

It is essential to select/assign appropriate background mu- sic (BGM) for each scene/cut when we edit a video or a slideshow of photos. However, it is a laborious task. Aim- ing to realise automatic BGM selection/assignment, we pro- pose a method to automatically assign emotion tag to various BGM. To realise this method, we need a model for classify- ing BGM. To build our model, we use a set of movie scene BGMs that a group of 14 users tagged with five (5) differ- ent sentiments: Love, Surprise, Joy, Sadness, and Fear. Af- ter confirming their agreements, we extracted the features of each audio file of our dataset. Using the machine-learning tool WEKA and the random forest algorithm, we built a model. Through a cross validation process, we evaluated our model and obtained an accuracy of 94% in prediction of the emotion in the BGM, demonstrating the effectiveness of the proposed approach.