A New Variant of Clustering-Based Normalization in Movie Recommendation System

Abstract

Several communities have experienced considerable growth in the Recommender System (RS) in recent years. Researchers are intrigued by it because of the recent expansion of several businesses engaged in E-commerce, and online video providers like Netflix, YouTube, Hotstar, etc. The collaborative filtering-based RS system strives to recommend to the viewers such movies or videos on their prior history. Typically, a rating matrix is used to represent these data. These ratings are not consistent though some users’ reviews are harsh, and others are liberal. Because of this, the RS is unable to recommend tailored movies to demanding viewers. This paper presents a collaborative filtering recommendation system that utilizes movie clustering and normalization to address the aforementioned problem. Using the average user rating for each movie and the number of users who evaluated each movie in the first stage, the suggested technique clusters the movies. In the first phase, it makes clusters of similar movies using Jaccard similarity then the RS calculates the normalized user count for each movie is determined using min-max normalization, Next, the users' scaled average ratings within a certain range are calculated. When the rating matrix creates user rating predictions during the assessment phase, it is divided into training and testing rating matrices.