Multi-source based movie recommendation with ratings and the side information

Many recommender systems are built based on the user ratings or user interaction data collected by the content or item providers. However, one data source may only provide limited information about the items. It would be helpful, if information about the candidate items could be retrieved from multiple data sources. In this work, a movie recommender system is designed relying on a variety of data sources that provide different types of user feedbacks on movies, including the MovieLens and Netflix rating data, YouTube movie trailer data, and movie-related tweets from Twitter. The feedbacks on movie trailers such as likes, comments, and tweets can be considered as the side information of the movies. They can be represented as movie features and then integrated with the movie ratings. Or, some of them (e.g., sentiments of the comments) can be represented as the implicit rating matrix and then integrated with the explicit ratings. The experiment shows that the inclusion of the trailer data improves the recommendation accuracy, and the most accurate result is achieved when all the feedback data is combined as the movie features.