A cross-modal cognitive reasoning framework for archival studies based on deep multimodal learning

With the development of artificial intelligence and big data technologies, archival studies have gradually expanded from traditional text analysis to the integrated use of multimodal data. However, existing methods still predominantly rely on a single textual modality, lacking a cross-modal cognitive framework that can effectively integrate and reason with multi-source information. This limitation restricts the in-depth analysis and application of archival resources. To address this, this paper proposes a cross-modal cognitive reasoning framework based on deep multimodal neural networks, enhancing the comprehensive analysis and management capabilities of archival resources by integrating text, images, and audio data. This framework combines natural language processing, computer vision, and speech recognition technologies to achieve semantic fusion and collaborative reasoning of multimodal information. Experiments on the Archival-MultiModal and DocVQA datasets show that the proposed framework achieves an accuracy of 91.7%, a 2.6% improvement over the best baseline model (CLIP), with inference time reduced by approximately 48%. The results demonstrate that this framework exhibits excellent performance and robustness in handling complex, multi-modal archival data. This research provides a new approach to cross-modal cognitive tasks in archival studies and offers valuable insights for advancing the intelligent development of archival resource management.