A systematic modelling of recommender system using derivative imitation of SVD and regression

Collaborative filtering has been extensively used in recommender systems, and its adoption is aided by the abundance of actual and trustworthy user data available in the era of big-data. However, as consumers' knowledge of information security grows, these data either disappear or deteriorate in quality. Matrix factorization approach - SVD (Singular Value Decomposition) is often utilised in Collaborative Filtering (CF) systems. It adds bias information about users and products and is accomplished with the use of algebraic feature extraction. Due to the inclusion of implicit feedback information, the SVDpp, derivative model of SVD obtains a higher prediction accuracy. But it also struggles to produce accurate predictions in case of highly sparse datasets. In this paper, a hybrid model blending the functionality of SVDpp and regression technique is proposed. The performance of proposed hybrid model is compared with other baseline models such as: KNNBasic, KNN Means, KNN ZScore, SVD, Non-negative Matrix Factorization (NMF), SlopeOne and Co-Clustering technique. The performance testing is done on MovieLens-100K dataset. The proposed model outperformed the baseline approaches with relatively better significance in terms of prediction accuracy.