Optimization and Application of Deep Learning Algorithms in Personalized Information Service Recommendation

This paper aims to optimize the library book recommendation system by using DNN (Deep Neural Networks) and UBCF (user-based collaborative filtering) methods, design a more efficient HR (hybrid recommendation) model, and improve the performance of the recommendation system and user experience. The current book recommendation system still has obvious deficiencies in dealing with the diversity of user interests and data sparsity, which affects the accuracy of recommendations and user satisfaction. This paper combines the feature extraction capabilities of deep learning (DL) with the user behavior analysis of CF (collaborative filtering) to improve the personalization and accuracy of recommendations and solve the bottleneck problem in traditional recommendation systems. This paper compares the performance of the HR model, DNN model and UBCF model in terms of recommendation accuracy, F1 value and Top-N accuracy. The results show that the average recommendation accuracy of the HR model in 7 book categories is 92.3%, which is significantly higher than the 89.3% of the DNN model and the 84.2% of the UBCF model. The average F1 value of the HR model is 0.92, which is better than the 0.88 of the DNN model and the 0.84 of the UBCF method model; in the Top-10 accuracy, the HR model reaches 93.3%, which is a significant improvement over other models. This shows that the HR model overcomes the limitations of a single model by combining the neighborhood relationship of CF and the feature mining capabilities of DL. It has significant advantages in both recommendation accuracy and applicability, providing new ideas for the optimization of book recommendation systems, and has significant practical significance and application potential.