Contents Popularity Prediction by Vector Representation Learned from User Action History

摘要

The anime and manga industry is important in Japan, and its popularity has been increasing overseas in recent years. Under such circumstances, predicting the popularity of media contents is important for content holding companies. Popularity prediction research has, so far, rarely considered the multi-faceted information of media contents based on consumer preferences. In this study, we extracted users' preferences from Wikipedia and obtained a vector representation with multifaceted content information. We qualitatively analyzed learned vector representations and showed that accuracy is improved by 2 to 3% in a popularity prediction task.