A Matrix Factorization Algorithm for Movie Recommendation

In this paper, we build various movie recommender system ranging from simple recommender systems to hybrid recommender system. We used content as well as a collaborative filtering technique. To establish the similarity relation between users or movies we have used cosine similarity measures. We analyse the used of user-user similarity matrix as well as a movie-movie matrix in building our recommender system which plays a very crucial role in determining the accuracy rate and computational time consumption. In content-based recommender system, for deriving similarities between movies from their description corpus we have used the Term frequency-Inverse document frequency measures. Next we move on to build collaborative filtering using various models like K-nearest neighbours, K-nearest neighbours with mean, Singular value decomposition and Singular value decomposition ++ and compare them in terms of their accuracy in Root Mean Square Error with the least time consumption. Prior to all, we utilize the Internet Movie Database weighted rating formula to build a simple recommender system.