Fully content-based movie recommender system with feature extraction using neural network

In recent years, movie industry is getting more and more prosperous. There are hundreds of movies released every year. However, it is difficult to notice the releasing of every movie, not to mention actually seeing it. Therefore, movie recommender system has become more and more popular as a research topic. Among a variety of movie recommender systems, content-based methods always ring a bell when it comes to recommending new movies. Content-based method uses the content of the movie as input so that it does not suffer from the “cold-start” problem. In this paper, we propose the Fully Content-based Movie Recommender System (FCMR) to recommend movies to users. The proposed method trains a neural network model, Word2Vec CBOW, with content information (e.g., cast, crew, etc.) as the training data to obtain vector form features of each element, and then take advantage of the linear relationship of learned feature to calculate the similarity between each movie. In the end, the proposed FCMR recommends movies based on the similarity. The experiments are conducted on a massive real world dataset, and the intuition behind our proposed method has been proven by the experiment results.