| The popularization of knowledge of potential safety hazards investigation in logistics enterprises requires users to screen valuable knowledge information of potential safety hazards investigation,which is not enough to meet the needs of ordinary logistics enterprise users to obtain knowledge directly.This paper constructs a knowledge graph for logistics enterprise users and integrates it into the safety production hidden danger troubleshooting question-and-answer(Q&A)system,so that logistics enterprise users can directly obtain the relevant knowledge of safety hidden danger troubleshooting,so that the Q&A system can answer users’ natural language questions in a direct and accurate way,so as to improve the effect of users’ interaction.The main work accomplished includes:(1)A structured parsing model of hidden danger description text based on ERNIE-CRF-ESL is proposed.According to the characteristics of security risk description text,this paper first designs a sequence annotation method suitable for security risk description text,and proposes word vector feature extraction based on the enhanced representation from knowledge integration(Ernie)model.On this basis,by fusing conditional random field(CRF)module and information extraction(ESL)module.The specific triple knowledge information is extracted according to 11 matching patterns,and the joint learning model based on ERNIE-CRF-ESL is established.Experiments are carried out on the collected security risk description text.The experimental results show that the accuracy of the proposed model in the text structured parsing task is 65.1%.(2)The construction method of knowledge graph of safety production hidden danger troubleshooting in logistics enterprises based on ontology modeling is proposed.According to the characteristics of many types of hidden danger investigation data of logistics enterprises,this paper carries out the standardized treatment of hidden danger sources based on ernie-crf model.Using the ability problem to define the focus and scope of the knowledge graph,describe the scope of the ontology in detail,and design the ontology model in the field of safety production hidden danger troubleshooting.The first knowledge graph of potential safety hazards investigation for logistics enterprises is constructed by linking the knowledge with the data of hidden danger investigation of logistics enterprises.Neo4j graph database is used to store and visualize the knowledge graph,which provides data support support for the development of the Q&A system.(3)An intelligent question answering model based on retrieval is proposed.In order to transfer knowledge information to users more accurately,this paper abstracts user problems through word slot design in the user problem preprocessing module;In the design of abstract problem understanding module,through the improved convolution size on adjacent words of Text-CNN,the intention recognition model can obtain the complete intention characteristics of abstract problems;In the design of the answer retrieval module,the user’s words and the user’s company information are transformed into idiom semantic feature vectors through Bert-as-service,and the hierarchical navigable small world(HNSW)algorithm is used to index the logistics enterprise and hidden danger information,and search the candidate answers for similarity;Finally,combined with the user intention and the candidate answers,the answer generation module encapsulates the final answer by designing different query statements and feeds back to the user.This paper analyzes the needs of users of logistics enterprises for hidden danger troubleshooting Q&A system.The three-tier architecture design method is used to design the system architecture.Combined with the knowledge graph of hidden danger investigation of logistics enterprises and the intelligent Q&A model based on retrieval,the construction of hidden danger investigation knowledge Q&A system of logistics enterprises is developed.The system can accurately answer 8 kinds of knowledge questions related to safety production that users may be eager to know at the same time.Through the system test,it is verified that the Q&A system can accurately and quickly calculate the answers related to users’ questions. |
