Construction of Knowledge Graph in Clothing Commodity Field Based on BERT-Lattice-Att-CRF Model

Consumers rely heavily on diversified information such as user reviews and clothing knowledge when shopping for clothing online. As a tool for effectively integrating and utilizing massive amounts of data, knowledge graphs are currently primarily used in the field of clothing for product recommendations, but there is little research on clothing knowledge graphs that integrate clothing attributes, user reviews, and domain knowledge. Therefore, this paper constructs a clothing domain knowledge graph that integrates multi-dimensional information. First, we designed a construction process for the clothing domain knowledge graph and proposed an ontology construction method. We mined structured knowledge from multi-source data, and addressed the issues of blurred entity boundaries, nested entities, and polysemy in Chinese clothing domain texts by proposing an entity recognition model based on BERT- Lattice-Att-CRF. The introduction of the BERT model and self-attention mechanism improved the model's semantic representation ability and long-distance semantic information processing ability. Experiments showed that this model significantly improved the accuracy of clothing entity recognition. Finally, Neo4j was used to visualize the knowledge graph. This paper provides new ideas and methods for the construction of clothing knowledge graphs.