With the rapid development of the digital entertainment industry, accurately predicting the popularity of film and television content is of great significance for optimizing recommendation systems and making business decisions on content platforms. This study proposes a comprehensive predictive framework that integrates user behavior features, movie content features, and collaborative filtering information to construct multiple machine learning models for predicting movie ratings and popularity. The experiment was conducted on the MovieLens dataset, comparing traditional machine learning methods (linear regression, random forest, XGBoost) with deep learning approaches (multilayer perceptron), and further enhanced predictive performance through ensemble learning strategies. The research results indicate that the XGBoost model achieved the best performance in rating prediction tasks (RMSE=0.862), while the ensemble model reduced prediction error by 8.3%. Through SHAP value analysis, the study identified that the historical average rating of movies, user rating behavior patterns, and movie genres are the three most critical factors that affect prediction. This study provides empirical support and methodological guidance for the optimization of film and television recommendation systems.
Yutong Liu (Mon,) studied this question.