Research on digital classification and intelligent retrieval of Chinese language and cultural resources based on natural language processing

Chinese language and cultural resources are the core carrier of Chinese civilization, covering ancient books, dialects, folk records and online literature. However, the current management of cultural resources faces many challenges, such as the low efficiency of digitalization of ancient books, the dependence on manual phonetic labeling of dialects, and the lack of semantic analysis of classification standards, which leads to low resource utilization. This study proposes a framework for digital classification and intelligent retrieval of Chinese language and cultural resources based on natural language processing (NLP), aiming at improving the automation and intelligence level of cultural resources management by integrating cultural ontology knowledge and multimodal features. In the research, a multimodal data preprocessing and feature fusion module is constructed, and the OCR-ERNIE model is used to process the text mode, Wav2Vec 2.0 is used to process the audio mode, and the CLIP-ViT model is used to process the image mode, and the graph embedding algorithm is used to generate the cultural term map. On this basis, the classification model of cultural semantic enhancement is designed, and the classification effect is improved by combining the classification loss and semantic relevance goals through the multi-task joint training framework. At the same time, a cross-modal comparative retrieval system is constructed, which maps multi-modal data into a shared semantic space through a comparative learning framework, and introduces a cultural semantic weighting mechanism to enhance the cultural explanatory power and accuracy of retrieval results. In addition, the optimization strategy of user interaction is put forward, including active learning annotation method and feedback mechanism based on multi-arm slot machine to improve the usability and retrieval effect of the system. In the experimental part, the multimodal cultural resource data set CulData-1.0 is integrated, and the contribution of each module to the performance is verified by ablation experiments. The results show that CulNet model is superior to baseline model in text classification, image classification and cross-modal retrieval tasks, especially in fine-grained classification and multi-modal association modeling.