A Combined Approach For Collaborative Filtering Based Recommender Systems with Matrix Factorisation and Outlier Detection

Abstract

ABSTRACT Recommender system is a data sifting tool that can recommend items that can be of interest to the user. Collaborative filtering (CF) makes recommendations based on the ratings the users give to items. But noisy or inaccurate ratings reduce the quality of the recommendations. In spite of extensive studies carried on CF-based recommenders, a robust recommender to handle outlier in dataset is a challenging problem. In this study, a Factor wise Matrix Factorisation model (FWMF) is proposed for the prediction of item rating in recommender systems. To further strengthen the proposed FWMF model, a meta learning model that combines density-based outlier detection and bagging outlier detection is proposed to detect outliers. The outliers predicted are eliminated, and a comparative analysis is carried with FWMF to find the effect of outliers in making recommendations. The experiments were analysed with various error metrics conducted on benchmark dataset show that the proposed outlier extent recommendation model outperforms the conventional CF-based systems.