Enhancing Movie Recommender System

Abstract

Recommender Systems are used by customers to buy items more efficiently. Business is also benefited simultaneously. There are various approaches of Recommender Systems, like: (1) Collaborative Filtering (2) Content based Filtering and (3) Hybrid Filtering. In Collaborative Recommender System, ratings of the most similar users (in user based collaborative filtering) or items (in item based collaborative filtering) are used to predict the rating of a new item. In Content Based Filtering, user profile is constructed based on the content of the items liked by the user in the past and then based on similarity between user and item profile, prediction is made. Hybrid Filtering combines collaborative and content based approach. In this dissertation, we focus on movie recommendation task. We propose a new hybrid approach which combines usage, tag and other content data of items. We model prediction task as classification problem where our aim is to predict whether the item will be liked or disliked by the user. In our work, we propose item based Recommender which combines usage, tag and movie specific data such as genres, star cast and directors to improve the accuracy of the Recommender System. We have tested our approach using Hetrec2011-movielens-2k dataset. We have used Accuracy, Precision, Recall and Fmeasure to evaluate the performance of our proposed algorithm.