Research On The Construction Of Non-standard Product Knowledge Graph For Human-computer Dialogue

Non-standard products are not products manufactured by uniform parameter standards,but products manufactured by manufacturers to meet the needs of customers and markets.The knowledge graph of non-standard products can help the transformation and upgrading of nonstandard products to intelligence.The difficulty in building a knowledge graph of non-standard products for human-machine dialogue is that non-standard products do not have uniform attributes.Therefore,different knowledge ontology systems are required for different types of products,and the text corpus has features such as multiple levels of relationship semantics and complex data structure.Knowledge extraction needs to be carried out in combination with the characteristics of the domain corpus.Based on the above,this paper takes human-machine dialogue as the application scenario,taking the virtual anchor of Nankang Furniture as an example,and builds a knowledge graph of non-standard products according to the business needs of the industry.The specific research contents are as follows:(1)A named entity recognition model based on non Bert-Bil STM-CRF is proposed.Firstly,aiming at the application scenario of human-machine dialogue,the domain knowledge ontology of non-standard products is modeled by literature analysis and other methods;Secondly,unstructured data is obtained by means of crawlers,and multi-source data fusion is carried out to generate the original corpus;Finally,the entity extraction based on Bert-Bil STM-CRF named entity recognition model is completed.The word-level embedded vector representation of the data is obtained by using the deep semantic information provided by Bert,a pre-training language model,and the sequence semantic feature matrix is generated by combining the context features of the Bit-Bil STM-captured corpus.Finally,the sequence feature matrix is input into the CRF module for decoding,and the output prediction tag sequence is obtained,and the named entity recognition model is compared.(2)A relationship extraction method based on Bi LSTM Attention+Position is proposed.Based on the analysis of the characteristics of non-standard product corpus data,the word embedding vector and position feature embedding vector are fused at the embedding layer to form a joint feature embedding vector,which is input into the Bi LSTM Attention+Position model for relationship extraction.In addition,this article inputs non-standard product data into a Bi LSTM Attention relational extraction model that does not fuse position feature vectors for comparative experiments.The experiments show that the Bi LSTM Attention+Position relational extraction model has better results and significantly improves evaluation indicators.(3)Taking Nankang furniture as an example,the construction and application research of non-standard product knowledge graph is proposed.Firstly,taking the virtual anchor of Nankang Furniture as an example,the research on the construction of non-standard product knowledge graph oriented to human-machine dialogue is completed;Secondly,it introduces the method of importing the knowledge base into the Neo4 j diagram database,using Neo4 j to store and visualize the knowledge base,and displaying the context of non-standard product domain knowledge in a graphical manner;In addition,Cypher statements are used to provide knowledge services for front-end interactions;Finally,in order to better serve the needs of human-machine dialogue,a knowledge inference based on the translation distance Trans E inference model is proposed to supplement and update the knowledge graph.Research on the Construction of Knowledge Map for Non standard Products Oriented to Human Machine Dialogue can obtain knowledge triplets from non standard product data and complete the entire process of building the knowledge map.Taking the virtual anchor of Nankang Furniture as an example,this article provides a solution for building a knowledge map of nonstandard products oriented to human-machine dialogue,providing rich and accurate low-level corpus support for downstream applications such as human-machine dialogue,and has practical application value.