Research on Book Borrowing Prediction Model Based on Transformer

Abstract

With the increasing demand for intelligent library services, accurately predicting user borrowing behavior has become a key issue for optimizing resource allocation and improving service quality. Traditional prediction methods have significant shortcomings in long sequence modeling, feature interaction, and cold start scenarios. This paper proposes a book borrowing prediction model based on the Transformer architecture. It dynamically captures the long-term dependency of user behavior through the self-attention mechanism and combines book metadata and temporal context features to achieve multi-dimensional information fusion. Experiments show that it significantly outperforms the baseline model. The study further explores the application potential of the model in dynamic inventory management, personalized recommendation, and other scenarios, providing theoretical support and technical solutions for the intelligent transformation of libraries.