CKComCF: Canopy–K-means Clustering based Combined Collaborative Filtering

Abstract

In recent times, big data is revolutionizing every aspect of human lives. It has attracted more attention in academia and industry. To quickly obtain information, information filtering systems are essential. In the e-commerce industry, the recommendation system (RS) to predict the preferences of people has been prevalent over the few years. Collaborative filtering (CF) is one of the most conventional algorithms of RS. However, CF suffers from data sparsity and scalability issues. Thus, we propose Canopy–K-means Clustering-based Combined Collaborative Filtering (CKComCF) to solve the challenge of data sparsity and scalability. In particular, the prediction outcomes of user-based CF (UbCF) and item-based CF (IbCF) are integrated using a weighting approach, which is based on the root-mean-square error (RMSE) minimization. Experiment results based on two real-life datasets of MovieLens and Netflix Prize demonstrate that the proposed RMSE-minimization method outperforms the traditional CF methods, improving the accuracy by 64.24% (UbCF with MovieLens) and 13.72% (IbCF with Netflix Prize). The proposed CKComCF model outperforms the existing improved CF method, reducing the calculation time by 41.84% (MovieLens) and 64.77% (Netflix Prize).