Movie Recommendation System Employing Latent Graph Features in Extremely Randomized Trees

E-commerce development has made recommender systems compulsory to prune large amount of data so that users can be directed toward the items that are best suitable to them. Mainly, recommender systems are classified into 1) Content Based approach, 2) Collaborative Filtering approach and 3) Hybrid approaches; which in general relies on various features of users and items requiring a detailed and organized information about them. In many situations, it is next to impossible to gather such a massive data in systematic form. Therefore, our main aim in this paper is to implement an efficient recommendation system which requires only three attributes: User-Id, Item-Id and Rating given by User to Item. We propose a completely different approach, which generates latent graph features from User-Item rating links and uses them for recommendation. We apply it for predicting Movie Ratings, using just User-ids, Movie-ids, rating and not taking in consideration factors such as actors, genres, director, etc. In the experimental section, numerical results are provided for the proposed algorithm. MovieLens dataset is used for the experimental purpose. Results show that it outperforms existing techniques in current literature.