Online Movie Recommender System

Abstract

The rapid growth of internet has brought about new trends in information management and retrieval, one of which is Recommender Systems (RS). RS were developed to solve the problem of information overload for users, and help providers offer the right products and services to the right customers. In this thesis, two popular techniques for RS – Collaborative Filtering (CF) and Content-based Filtering (CBF) are introduced, studied and applied in the MovieLens online movie datasets. The two main divisions of CF are also analyzed and compared to find the best practice for this problem. In order to compute the predicted ratings of a user for an item, an CFRS requires the similarity matrix between users, and a nearest neighbors’ predictions algorithm, while the CBFRS demands for item profiles of movies. For the final result, the system proposes an optimal number of neighbors for the CFRS problem, and the method with the best prediction accuracy. Keywords: Recommender Systems, User-User Collaborative Filtering, Item-Item Collaborative Filtering, Content-based Filtering, Predicted Rating