Research on intelligent recommendation system of library information resources based on deep learning

In this paper, an intelligent recommendation system of library information resources based on deep learning is studied. Multilayer Perceptron (MLP) is adopted as the core recommendation model, and a double-tower structure is designed to improve the calculation efficiency. The user-side network extracts features from user data, and the book-side network extracts features from resource data, and realizes personalized recommendation by calculating the similarity between them. In addition, the system also introduces explicit feature crossover to enhance the ability to capture combined features, and uses FAISS library to optimize the recommendation generation time. In the performance evaluation of the system, a six-month borrowing record data set of a university library is used for experiments, and user-based collaborative filtering (CF), TF-IDF-based text similarity recommendation (Content-based) and Hybrid model are compared. The results show that the proposed two-tower MLP model (DeepRec) is significantly superior to other models in recommendation accuracy (Precision@10, Recall@10, NDCG@10, AUC-ROC), especially in the cold start scenario of new users, NDCG@10 is improved by 78%. The system shows good stability in high concurrency scenario, and the response time of P99 is stable within 60ms at 500 concurrency, which meets the production requirements. At the same time, the system supports horizontal expansion, and the throughput increases approximately linearly with the increase of the number of nodes, showing good scalability. The intelligent recommendation system of library information resources based on deep learning designed in this study is excellent in recommendation accuracy, cold start ability and system stability, which can effectively improve the recommendation effect of library information resources and provide users with more personalized and efficient services.