Research And Implementation Of Question Answering System Based On Brand Index Knowledge Graph

Brand building is an important work for China to become stronger.General Secretary Xi Jinping's speech on Three Transformations,Opinions on Giving Full Play to the Leading Role of Brands to Promote the Upgrading of Supply and Demand Structure issued by General Office of the State Council and the 13 th Five-year Plan for Quality Brand Promotion issued by AQSIQ(General Administration of Quality Supervision,Inspection and Quarantine of the People's Republic of China)determine the national brand strategy.Tianjin Brand Index and Evaluation Method which formulated by Tianjin is an important part of the Tianjin brand building.Based on knowledge graph and question answering system technology,this paper not only constructs the Brand Index Knowledge Graph,but also researches and develops the Brand Index Question Answering System,which can answer the questions related to the brand index platform.The system mainly includes five modules: knowledge graph construction,system interaction,question generation,question analysis,and answer generation.To solve the problem of entity labeling error and multi-entity duplicate name in the traditional question analysis model,the word embedding matching model is added to the traditional question analysis model,so that the entity with the shortest distance from the question can be found in the knowledge graph as the main entity of the question.To solve the disproportion problem of the brand index question answer dataset on different types of sample distribution,this paper constructs the automatic question model,the model uses the Transformer Model in the question generated task of the knowledge graph,meanwhile,the answer information is added to the input of the model which combines with the Copy Mechanism,the model can improve the pertinence and readability of generation question,so as to achieve the data enhancement of the brand index question answer dataset.This paper has taken on the contrast experiment,through the analysis of the experimental results,in the data-enhanced dataset,the question analysis model can obtain the best accuracy of 93.62%,compared with the original dataset,the accuracy rates increases by 1.31%,this paper also develops the system which can meet the user's basic query requirements for the brand index platform and enable user to interact with the system in natural language to improve the user's query experience.