A Collaborative Filtering Recommendation System for Rating Prediction

Abstract

Recommendation system is a subclass of information filtering system to help users find relevant items of interest from a large set of possible selections. Model-based collaborative filtering utilized the ratings of the user–item matrix dataset to generate a prediction. Essentially, this type of intelligent system plays a critical role in e-commerce, social network, and popular domains increasingly. In this research work, we present the comparison of the two widely used efficient techniques such as Biased Matrix Factorization and a regular Matrix Factorization, both using Stochastic Gradient Descent (SGD). We have conducted experiments on two real-world public datasets: Book Crossing and Movie Lens 100 K and evaluated by two metrics such as Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Our experiments demonstrated that Biased Matrix Factorization used SGD technique results in a substantial increase in recommendation accuracy for rating prediction in experimental both datasets. Compute with a regular Matrix Factorization technique, Biased Matrix Factorization produced the reduction of the RMSE by 25.78% and MAE by 19.69% for Book Crossing dataset and RMSE by 19.69% and MAE by 14.08% for Movie Lens 100 K dataset. As expected when comparing the results of different datasets, Biased Matrix Factorization using SGD materialize less prediction error.