Movie Rating Prediction based on Data Mining and Machine Learning

Abstract

In personalized recommendation, accurately predicting the user’s rating of movies can significantly improve the quality of recommendations, enhance user experience and platform stickiness. In the analysis of the film industry, it is helpful for film market performance evaluation, audience preference analysis and content creation guidance; In the field of academic research, it is also one of the important benchmarks for testing the performance of data mining and machine learning models. This paper systematically reviews the research progress of film rating prediction based on data mining and machine learning. This paper first reviews the traditional prediction methods relying on metadata and collaborative filtering, and analyzes their advantages and limitations. Then, the sentiment feature extraction and modeling method based on text comments is discussed, and the role of sentiment information in improving the interpretability and accuracy of prediction is expounded. Furthermore, the multimodal fusion strategy of integrating metadata, text, vision, audio and other multi-source information is discussed, and the technical characteristics and research status of different fusion levels are summarized. This paper also summarizes the common datasets and evaluation indicators in this field, and points out that there are still deficiencies in data integrity, modal fusion depth and interpretability. Finally, this paper looks forward to future research in the directions of deep multimodal interaction, cold start mitigation, and model interpretability enhancement, in order to provide a systematic reference for researchers in related fields.