Improve the Quality of Recommender Systems based on Collaborative Filtering with Missing Data Imputation

Abstract

Recommender systems which are becoming increasingly popular and have been proven to boost cross-selling and customer satisfaction still need further refinement to improve their relevancy and correspond to the user’s need. One widely used method of recommender systems is Collaborative Filtering (CF). The CF method bases predictions for a user on feedback data from other users, so that user preferences can be formulated based on the feedback it provides. One of the prevalent obstacles to the CF method is that the feedback data is often incomplete because there is no compulsion for the user to provide feedback on the items they have consumed. Incomplete data will affect the quality of recommendation results. One way to address this is to use missing data imputation. In this study, a collaborative filtering method is proposed where missing data imputation is carried out first before commencing data training. From the experiment results using Amazon Customer Review dataset, it is shown that the CF method with Imputation Mean produces lower RMSE and MAE values.