MovieANN: A Hybrid Approach to Movie Recommender Systems Using Multi Layer Artificial Neural Networks
Sait Can Yücebaş · Journal of Advanced Research in Natural and Applied Sciences · 2019
The amount of data in World Wide Web is growing exponentially. Users are often lost in this vast ocean of data. In order to filter the valuable information from vast amount of data, recommendation systems are used. These systems are based on collaborative filtering, content based filtering and hybrid approaches. We combined collaborative and content-based filtering to build a hybrid movie recommendation system, MovieANN, based on neural network model. To make better recommendations in a collaborative approach, both user and movie clusters are formed. In addition to rating information, content information was also considered in the formation of the clusters. Clusters are formed according to K-Means and X-Means algorithms. Final clusters are chosen according to Davies-Bouldin Index and intra cluster distance. Homogeneity and density of the clusters are also considered. Movie and user clusters are mapped in the recommendation phase. The model is tested on a MoiveLens 1M dataset that consists of six thousand users, four thousand movies and one million ratings. Four clusters are formed to represent movie – user mappings and for each cluster, a recommendation model based on multi-layer neural network is constructed. The recommendation performance in terms of accuracy is 84.52%, 84.54% in terms of precision and 99.98% in terms of recall.