A collaborative filtering recommendation algorithm based on normalization approach

Recommender system (RS) has grown widely in various communities over the last few years. It creates curiosity among the researchers due to the recent growth of various commerce companies, especially Flipkart and Amazon. In collaborative filtering-based RS, the system aims to provide the users with their personalized items, which is based on the users' past history. In general, these observations are represented in the form of rating matrix. However, these ratings are not uniform as some user ratings are stringent and others are lenient. As a result, the RS is incompetent to suggest the personalized items to the stringent users. In this manuscript, we design a normalization-based collaborative filtering recommender to overcome the above problem. The proposed algorithm consists of two phases, namely designing and evaluating. In the first phase, the proposed algorithm finds the average user rating per item and counts the number of users purchased each item. Then it uses min-max normalization to find the normalized user count per item and scale the average ratings of users in a specified range. In the latter phase, the proposed algorithm divides the rating matrix into training and testing rating matrix, and predicts the users' ratings. We perform rigorous simulations using a large variety of users and items, and compare the results with a collaborative filtering-based RS using ten performance metrics to illustrate the efficacy of the proposed algorithm. Moreover, we evaluate the results through a statistical test, t-test and 95% confidence interval.