Hybrid Model Approaches Toward Movie Recommendation Systems and Their Comparisons

In the contemporary digital era, individuals face an overwhelming challenge of searching for personalized content from a seemingly infinite pool of available options such as books, videos, articles, and movies. Simultaneously, digital content providers strive to attract and retain as many users on their platforms as possible for extended periods. This has led to the evolution of recommender systems, which aim to enhance user engagement and satisfaction by recommending content based on their preferences and interests. These systems employ sophisticated algorithms and data mining techniques to analyze user behavior and feedback, enabling them to make personalized recommendations to users, thereby improving their overall experience. In this paper, an investigation is conducted in which different models in machine learning with their combinations are evaluated to evaluate which model could give the best accuracy with actual ratings given by users. The models used are Lasso regression, XGBoost, K-nearest neighbor, surprise baseline only, SVD, SVDpp, and their combinations. The proposed ensembled model of XGBoost + surprise baseline + surprise KNN baseline + SVD + SVDpp, with the minimum value of RMSE as 1.111410 and MAPE value as 39.261054, is our best model and can serve as a reference for any future recommender system.