New Custom Rating for Improving Recommendation System Performance

Abstract

Abstract Nowadays, the recommendation system still attracts many researchers. Various new businesses focusing on online marketing (E-Commerce) have emerged since the covid-19 pandemic. This phenomenon allows recommending items based on Collaborative Filtering Recommender System (CF) to enhance the shopping experience for users. An accurate CF depends on how similarity algorithms can find similar profile users correctly. The traditional similarity measures are based on the user-item rating matrix. In this paper, we employ four custom rating combine with new rating formula from the popularity of users, items, and its original rating which called called New Custom Rating (Ncr). Instead of focusing on original rating only, our Ncr tried to maximize the recommender system's performance by employing users' popularity and items to derive new ratings value. Using those formulas will make the new value rating more representative and the calculation of the similarity algorithm more precisely. Finally, the increased accuracy of the recommendation systems will be achieved. We conduct Ncr in four custom rating algorithms and four algorithms of the reccomender system in five public data sets. The experimental results demonstrate that Ncr can increase the recommendation system's accuracy by decreasing RMSE, MSE, and MAE and increasing FCP and Hit Rate. Ncr may improved accuracy of the reccomender system in various recommender systems algorithms by employing the popularity of users and items in rating.