Research On The Construction And Embedding Of Clothing Knowledge Graph

In recent years,with the rapid development of artificial intelligence technology and clothing e-commerce,new applications such as clothing product retrieval and personalized recommendation have received extensive attention,but a large amount of clothing data has also been generated,and traditional storage technology is not very good.Utilize and exploit its own value.As a cutting-edge artificial intelligence technology,knowledge graph can make full use of the massive clothing data generated in the big data environment,store and manage it in a structured way,effectively make full use of clothing data,and improve retrieval efficiency.Furthermore,in order to effectively utilize the established clothing knowledge graph,the knowledge embedding in the knowledge graph needs to be represented as a vector.Aiming at the above problems,this paper completes the construction of clothing knowledge graph based on clothing product attributes,and proposes a knowledge graph embedding model.The core content of this paper includes the following three parts:(1)Based on the attributes of clothing products,combined with the ontology concept and graph database technology,the knowledge graph construction in the clothing field is completed.Taking the clothing product data of the e-commerce website as the main data source,the data is obtained through the crawler technology,the clothing information extraction is completed with the help of the deep learning model based on BiLSTM+CRF,and the Protege tool is used to construct the clothing domain ontology and generate RDF data.After using the semantic plug-in to complete the mapping of the RDF triplet data to the graph data structure,the clothing data is stored through the Neo4j graph database to complete the construction of the clothing knowledge graph.(2)The convolutional neural network is applied to the knowledge graph embedding represented by ConvE to capture the interaction information of entities and relationships,but its standard convolution has insufficient ability to capture feature interaction information and low feature expression ability.Aiming at the problem of insufficient feature interaction ability,a knowledge graph embedding model based on improved Inception structure—InceE model is proposed.The model firstly replaces the standard convolution with hybrid dilated convolution to improve the ability to capture feature interaction information;secondly,it uses a residual network structure to reduce the loss of feature information.(3)The proposed knowledge graph embedding model based on the improved Inception structure is tested on three benchmark datasets Kinship,FB15k,WN18 and clothing datasets for link prediction tasks.Compared with other classic models,this model also achieves better results in MRR,Hit@1,Hit@3 and Hit@10 indicators,which verifies that the model has a stronger ability to capture feature interaction information and adaptability.