An Ensemble Weighted User-Based Collaborative Filtering Recommender System
Mahamudul Hasan, Rambabu Nalagandla · 2024
The Recommender System (RS) is widely used in predicting the future preferences of users. Recommending future preference from the set of previously recorded items is the main agenda of this system. Movie recommendation is a kind of RS that helps people to find their liked movies and build a custom taste profile based on the previously given rating. The taste profile gives an overview of other users’ about the rated movie. Based on this profile, users’ contacted persons can be suggested movie lists by the social platform. Collaborative Filtering (CF) learn previous records and estimates new preferences over them. In the CF-based recommendation system, some traditional approaches named Cosine, Jaccard, Mean Squared Deviation (MSD), Jaccard Mean Squared Deviation (JMSD), Pearson Correlation Coefficient (PCC), Constraint Pearson Correlation Coefficient (CPCC) are used to build users’ taste profile. The performances of the similarity measures are subjective. That is, no similarity measure works fine in all circumstances. Jaccard and Mean Squared Deviation uniformly give an approach to JMSD. Similarly, in this paper, a comprehensive method named “Ensemble Weighted approach” has been proposed which will be represented by the aggregated approach of some top baseline approaches. The final rating has been calculated by the prediction function. In the performance evaluation section, it is depicted that our approach outperforms in all circumstances and gives an accuracy that secures the best result.