Abstract:

[Purpose/Significance] As digital technologies such as 5G and generative AI become more prevalent in higher education, university libraries have evolved from traditional collections of books to ecosystems of cross-modal and multi-source resources, encompassing core collection resources, open-access resources, and user-generated content. However, the "resource silo" issue caused by heterogeneous resources and the mismatch between passive services and dynamic user scenarios in research and teaching remain unresolved. Existing studies lack integrated closed-loop mechanisms linking resources, scenarios, and users. This study aims to address these gaps by promoting libraries' transformation from "resource storage centers" to "proactive knowledge service centers." Its key innovation lies in constructing a scenario-driven three-dimensional collaborative model, which bridges the disconnect between resource integration and scenario adaptation, providing theoretical and practical support for intelligent library development. [Method/Process] Guided by ERG demand theory and context-aware computing, this study adopts a mixed-methods approach combining literature research, technical design, and case validation. A three-dimensional collaborative model of "Resource Integration - Scenario Adaptation - Smart Services" was proposed. For resource integration, a "three-dimensional integration + four-step fusion" framework was developed: standardized access via unified DCAT-AP/RDA metadata and multi-protocol gateways, associative reorganization through cross-modal semantic matching and knowledge graph aggregation, and hierarchical storage (hot/warm/cold tiers). The four-step fusion includes data preprocessing, modality conversion (ViT, Whisper-large, YOLOv8 models), feature fusion (attention mechanism + Transformer encoder), and knowledge generation (knowledge graphs, rule bases). An innovative five-dimensional dynamic scenario model (S=f(P,R,S,T,C)) quantifies user profiles, resource attributes, spatial locations, temporal contexts, and social connections for precise scenario identification. Technically, a "cloud-edge-device" architecture provides support, while a hierarchical service pathway (instant/in-depth/customized services) and a multi-dimensional evaluation system (resource/service/user dimensions) ensure closed-loop optimization. [Results/Conclusions] The model effectively achieves in-depth integration of multi-source cross-modal resources and precise scenario adaptation. Validated through typical applications - full-cycle research support and immersive teaching (VR ancient book restoration, MR anatomy demonstration) - it significantly enhances resource utilization efficiency and user experience, resolving the core pain point of resource-scenario disconnection. The model strongly supports libraries' transformation from passive resource supply to proactive knowledge services. Limitations include limited application of cross-modal technologies to virtual reality resources, insufficient coverage of management and social service scenarios, and the need for long-term validation of the evaluation system. Future research will deepen large-model-aided cross-modal fusion, expand scenario coverage, improve the evaluation system with third-party participation, and promote inter-university resource sharing to better support higher education development.

Key words: university libraries, multi-source resource integration, multimodal large model, cross-modal fusion, scenario-based application, smart services, knowledge graph