A Collaborative Filtering Movie Recommendation System Based on Users Correlation and Weighted K-Means with High Accuracy

A Recommendation system is a BI tool that it uses data mining methods for guiding and helping the user to select the best items based on her/his preferences and in the shortest time. Despite more than two decades of academic research on recommendation systems, this issue is still one of the most up-to-date research challenges. Recommendation systems save the users time, increase their satisfaction and their loyalty to sales sites and lead to the development of e-commerce, by personalizing the recommendations of goods or services to site users. Nowadays, recommendation systems have many applications in various sectors of e-commerce, especially in media products such as books, movies, and music. The famous e-commerce sites such as eBay, Amazon, and Netflix and domestic sites such as Digikala, Divar, and Filimo widely use recommendation systems. These systems use a variety of big data filtering methods to provide appropriate recommendations. The most important and widely used filtering method is collaborative filtering (CF). In this paper, we implement three CF recommender systems based on the correlation coefficient between users, selecting the optimal number of neighbors and calculating weighted scores for unwatched movies. The best method with the least error is selected as the desired model. We use Movielens ml-latest-small 100k research dataset with 9742 movies and 610 users as input. The results showed 3.29% less RMSE error compared with the latest research that has used the correlation method.