Digital Management and Service Optimization of Local Library Collections by Deep Learning Algorithms

Abstract

This study addressed the digital management and service optimization of local library collections. A KEMFA (Knowledge‐Enhanced Multi‐Field Attention) model was proposed for local document classification. The model incorporated external knowledge from a knowledge graph into multiple document fields, including titles, keywords, and abstracts, through a knowledge expansion strategy. A knowledge mask matrix was introduced to mitigate interference among different fields. Furthermore, a multi‐field masked attention mechanism was designed to integrate the enhanced knowledge with original textual features, enabling fine‐grained semantic representation and more accurate document classification. Experiments were conducted on three datasets, including TNEWS‐10 W (Ten Thousand Subsets of Today's Headlines), DUEE (DuEE Event Extraction), and a multi‐label dataset collected from Chinese first‐class libraries. On the TNEWS‐10 W dataset, KEMFA achieved an accuracy of 0.911, a Micro‐F1 score of 0.943, a retrieval relevance (RR) of 78.1%, a Topic Navigation Path Hit Rate (TNPHR) of 72.3%, and an inference time of approximately 19.8 ms. On the DUEE dataset, the model achieved an accuracy of 0.924 and a Micro‐F1 score of 0.956. On the library subject dataset, the Micro‐F1 score, RR, and TNPHR reached 0.812%, 78.6%, and 73.4%, respectively, representing improvements of 0.012%, 2.4%, and 3.3% over the best‐performing baseline model. Statistical t‐tests further confirmed that the improvements achieved by KEMFA in terms of Micro‐F1, RR, and TNPHR were statistically significant compared with all baseline models (p < 0.05). In addition, KEMFA maintained stable performance across different book categories. Ablation studies demonstrated that both the knowledge mask mechanism and the knowledge fusion module substantially enhanced multi‐domain semantic understanding and classification performance. Overall, the proposed KEMFA provided an effective solution for intelligent local document classification, topic navigation, and retrieval services, offering valuable insights for the digital transformation of library management.