Recommendation System Model Using Matrix Factorization Bias Terms and Convolutional Neural Network (CNN)

This research presents the development of a recommendation system that integrates matrix factorization (MF) with bias terms and Convolutional Neural Networks (CNN) to enhance recommendation accuracy. The Movielens-100k dataset was utilized in a series of experiments to evaluate the proposed model's performance. In the first experiment, the CNN MF Bias Terms model, configured with four convolutional layers and 128 latent vectors, outperformed other models, achieving HR@10 of 0.735 and NDCG@10 of 0.541. Subsequent experiments confirmed the superiority of this model over CNN MF and standard MF models, maintaining consistent accuracy. For instance, the average HR and NDCG for the Top 1, Top 3, Top 5, and Top 10 were 0.592 and 0.449, respectively, with an execution time of 15–17 minutes. These results indicate the model's capability in improving recommendation quality. Future research may explore integrating diverse data types, such as review text, images, and videos, and adopt more efficient evaluation methods to enhance model performance and reduce computation time.