The Method of Library User Profiling and Knowledge Demand Prediction Driven by Big Data

With the advancement of the 'Education Power' policy, China's investment in education is increasing annually. Libraries, as key knowledge service hubs, face the dual challenge of diversifying user demands and providing precise resources. According to the 2023 National Education Development Statistical Bulletin, annual university library visits have exceeded 4.2 billion, with 79.8% being digital resources. User demands are stratified: basic education users seek structured subject knowledge, while higher education users focus on interdisciplinary trends. As library resources expand from paper to digital forms, user profiling models based on borrowing records have become less accurate, with accuracy dropping below 40%. To address this, the 14th Five-Year Plan calls for a paradigm shift towards ‘resources adapting to users’ through big data and deep learning. We propose a data-driven user profiling and knowledge demand prediction method defined as a sequence-to-sequence problem. The input data includes user behavior variables (e.g., visits, borrowing frequency), while the target sequence covers identity, interests, and knowledge needs. Our model includes three parts: Triple Encoding Mechanism, Interactive Causal Feature Extraction, and Interactive Multilayer Perceptron. Experimental results show our method outperforms models like SVM, Transformer, and LSTM in reducing prediction errors, achieving the lowest MSE and MAE values. This approach enhances resource utilization, reader satisfaction, and offers valuable insights for the development of smart libraries.