Movie Recommender System

Abstract

Online shopping is a trend and a way to go these days for buying many different kinds of products. Typically, before buying any product, customer sees historical ratings received by the product and then makes a conscious decision. This scenario is also applicable to a movie. User relies on various ratings and reviews given by other users before deciding about watching a movie. This form of decision making is useful but relies on general senses of mass. It does not consider individual’s taste and preferences. This dissertation aims to fill in this gap.

User-based and Item-based collaborative filtering is exercised in this dissertation for personalized movie recommendation. Similarity between users and items is computed through different possible combinations and their impacts on prediction error is studied. A novel contribution of this dissertation is the fusion of user-based and item-based collaborative filtering to predict the rating. Precisely, approaches based on Genetic Algorithm (GA), Classification and Regression Tree (CART), Random Forest (RF), Linear Regression (LR) and Support Vector Regression (SVR) have been employed to address the fusion challenge. Results are encouraging and demonstrates the usefulness and superiority of fusion approaches.