Moreopt: A goal programming based movie recommender system

Abstract

Recommender systems suggest relevant items to users by acquiring user preferences and exploiting them to build a type of user model. The main purpose of such a system is to match the most suitable item for the constructed user model. And hence, finding similar items for user preferences is the most crucial point of any recommender system. The state-of-art recommender systems suffer from handling the data sparsity problem. For this reason, the proposed recommender system combines content information of movie features (cast, director, genre, etc.) with a collaborative filtering approach. The similarity scores of movie features are supplemented by a goal programming model in the content-based approach. Pearson correlation is selected as a collaborative filtering algorithm that predicts movies to satisfy user tastes considering the content-based similarity scores. MovieLens dataset is used for experimental setup and Mean Absolute Error is measured for the comparison of approaches. The best average MAE score is 0.736 when the evaluation includes 300 training users. Also, the fastest sub-task is the movie recommendation for users having 2.34 s running time. The proposed system outperforms the rest of the studies in the literature and the experiments show that the overall system performance is increased when the content information is augmented by the collaborative filtering approach.