In order to analyze the similarities and differences in the technical inheritance of modern Lingnan landscape paintings and Western landscape paintings, this paper proposes a knowledge graph construction method oriented to the multimodal attribute information of paintings. The EIEA model is adopted for multimodal entity alignment, and the unimodal-based entity embedding is obtained through multiple pre-trained models. An enhanced entity representation module is developed to mine the interaction information between paired entity nodes. The ERNIE model is utilized to complete the relationship extraction and knowledge fusion tasks to construct a multimodal knowledge graph. Crawler technology is used to obtain 600 entity data, and the reliability of the proposed knowledge graph construction method is proved through experiments. Comparison reveals that the average value of the degree of Lingnan landscape painting is 0.034 higher than that of Western landscape painting, and the difference between the point-in degree and the point-out degree is 0.003 higher than that of Western landscape painting, which is not only an important intermediary and integration node, but also an active participant in cross-cultural interaction of techniques in the technique transmission network. Western landscape painting, on the other hand, reflects more of the systemic and endogenous nature of technique inheritance, with a relatively low density of cross-cultural integration interactions.
