Research On Intelligent Question And Answer System For Hazardous Chemicals Based On Knowledge Graph

Hazardous chemicals play an important role in industrial production and scientific research,promoting China’s economic development and technological innovation.At the same time,safety accidents caused by explosion or leakage of hazardous chemicals also pose a great threat to people’s life and property safety.The Q&A system can quickly answer the questions raised by users in related industries,thus helping users to quickly get the information they need,thus reducing the time needed to deal with hazardous chemical accidents and reducing the impact of accidents.And knowledge graph is a form of structured knowledge representation describing entities,concepts,events and the relationships among them,which can help Q&A system link the entities mentioned in the questions to those in the knowledge graph,so as to answer the questions more accurately.Intent recognition and slot filling are two important tasks in question and answer systems.Intent recognition requires identifying the user’s intention or purpose expressed in the interaction.Slot filling requires extracting relevant information from the user’s utterance based on recognizing the user’s intention and filling it into predefined slots.Although good results have been achieved for the intention recognition and slot filling tasks with the development of deep learning techniques,only a few scholars have jointly modeled the two tasks,and the jointly modeled models ignore the impact of prediction errors of the former task on the latter task.Therefore,based on the above problems,this paper proposes a j oint model of fused intention recognition and slot-filling model to design a complete hazardous chemical Q&A system based on the already established hazardous chemical knowledge graph.The main research of this paper is as follows.1.In this paper,we build a knowledge graph of hazardous chemicals.Firstly,we use crawler technology to obtain hazardous chemicals data from the national official database,so as to clean and format the structured,semi-structured and unstructured data obtained,and extract the corresponding entities,relationships and attributes from them.For the data obtained from other channels,knowledge merging is used to fuse them into the original knowledge base,and finally the knowledge is stored in the graph database to establish a knowledge graph to provide data support for the answer retrieval of the Q&A system later.2.This paper constructs a j oint model of fusion intention recognition and slot bit stuffing model.On the basis of shared encoder,sentence level intention recognition is optimized to word level intention recognition.On the basis of contribution encoder,the neural network of intention recognition and slot bit stuffing task is optimized to two-way LSTM,so as to optimize the performance of the two tasks.Through comparative experiments and ablation experiments,good results are achieved on the experimental data set,Finally,the effectiveness of the proposed model was once again demonstrated by introducing a pre-trained model.3.Based on the constructed knowledge graph,this paper designs a matching Q&A system architecture,completes the construction of the hazardous chemical Q&A system,and demonstrates the interface of user questions,historical questions and knowledge graph visualization of the Q&A system.