The process of intelligent library construction is advancing, and readers’ demand for personalized information services is growing. Traditional recommender systems do not fully consider the temporal characteristics of users’ borrowing behavior, resulting in limited recommendation accuracy. Meanwhile, challenges such as user interest evolution and cold-start problem constrain the improvement of library service quality. OBJECTIVE: To construct an intelligent library personalized information recommendation system based on temporal data analysis in response to the problem of underutilization of temporal features in library personalized recommendation systems. Methods: Adopt convolutional neural network for local characterization of time-series data, and extract local features of user-item scoring matrix through normalization and similarity calculation; design personalized recommendation model based on BiLSTM, integrate Embedding layer and multilayer perceptron, and extract features of readers’ borrowing preference using bidirectional long and short-term memory network; construct recommendation system with B/S architecture, adopt MVC three-layer architecture and J2EE technology to realize the system functions. RESULTS: The experiments show that in the sparsity interval of 0.7-0.9, the accuracy of CNN local similarity prediction is higher than the Euclidean distance and Pearson correlation coefficient methods; the BiLSTM model achieves the optimal performance at the learning rate of 0.001, 2-layer network, and batch size of 256, with the MAE value of 0.787; compared with the UserCF, ItemCF and ConvMF algorithms, the proposed algorithm performs optimally in 10 experiments. CONCLUSION: The recommender system based on temporal data analysis effectively improves book recommendation accuracy and provides technical support for personalized library services.
