This paper explores the method of constructing a genetic knowledge map of Guanzhong residential culture by integrating multimodal data in order to reorganize the knowledge of Guanzhong residential culture. The study firstly proposes four entity types in terms of the content structure of Guanzhong residents’ culture: deity culture, Confucianism culture, dwelling culture and aesthetic concept. Then the identification and extraction process of Guanzhong residential culture genes was designed, and the visual features of Guanzhong residents’ culture were extracted based on the selected text data and picture data. Finally, Bert-wwm-ext-BiLSTM-Attention-CRF and BiLSTM-PCNN-Attention are selected as the entity recognition model and the relationship extraction model, respectively, to construct the Guanzhong residential culture gene knowledge graph. The knowledge graph construction method selected in this paper is validated, and the results of multi-model comparison experiments show that the entity recognition accuracy of Bert-wwm-ext-BiLSTM-Attention-CRF reaches more than 70%, and the relationship extraction performance of BiLSTM-PCNN-Attention has a better performance advantage among all the compared models, and this paper Both models designed are suitable for building Guanzhong residential culture gene knowledge map.
