Presenting a Semantic Orientation Based Method for Multi-Label Classification of Movies Content Using Their Subtitle Texts

Understanding movies content and their genre, is always a complex and open issue to researchers. Several activities have been carried out by researchers to find out movies content. Most of the activities conducted in this area have been using audio processing or video processing. Recently a group of researchers have proposed the idea of using movies subtitle texts to understand movies content and considered text processing faster and easier than audio and image processing. In this paper a semantic orientation based method is presented for genre classification in multi-label data of movies subtitles. To do this, a feature extraction method is presented to extract unique features of each genre. Then a method is presented, in which with calculation of a subtitles semantic orientation to each genre, subtitles genres are predicted. Finally, using association rule mining methods, the relationship between genres in raw data is discovered and using these rules, predicted genres have been modified. Obtained results indicate significant improvement of proposed method compared to previous methods.