Research And Implementation Of Intelligent Question Answering System Based On Knowledge Map

With the rapid development of the Internet,online shopping has become one of the main forms of consumption for people,while e-commerce platforms have developed intelligent customer service Q&A systems to better serve online shopping users and merchants.With the rapid growth of the variety of products and the number of online shoppers,it is increasingly difficult for online shoppers and merchants to obtain real and effective product evaluation information from the massive amount of product review information.The existing intelligent customer service Q&A system cannot meet such personalized needs of online shoppers and e-commerce merchants,so it is necessary to build an intelligent Q&A system for the e-commerce field.Intelligent Q&A systems involve numerous technologies such as named entity recognition,relationship extraction,knowledge base construction and database retrieval.In the fine-grained product evaluation of e-commerce,the extraction of entities and their aspects evaluation is a difficult problem.To address the above needs,this paper builds a knowledge graph-based intelligent question and answer system based on e-commerce product review data.The main research work is as follows.(1)In response to the current situation that there is no publicly available entity recognition dataset for product reviews in the e-commerce domain,this paper adopts the Scrapy crawler framework to crawl the laptop computer review data on the Jingdong platform,and after processing the data,it uses the BIO tagging strategy for manual tagging to build an entity recognition dataset for product reviews in the e-commerce domain.(2)In e-commerce product review entity recognition,a named entity recognition method based on the BERT-Bi GRU-MHAT-CRF model is designed and its effectiveness is verified through experiments.The method firstly vectorizes the laptop review data by BERT and uses it as the input of Bi GRU;then,it automatically learns the text features in the laptop review data by Bi GRU network and uses the multi-headed attention mechanism for feature reinforcement;finally,it uses CRF for label prediction to identify the entity classes in the laptop review data.(3)In terms of entity relationship extraction,a lexicon and rule-based approach is used for entity relationship extraction.First,based on the entities identified from the e-commerce product review data and the existing evaluation word lexicon,a lexicon of evaluation objects and evaluation words is constructed for the laptop review aspect;then,the rulebased extraction of evaluation object-evaluation word relationships is combined with the lexicon to obtain entity relationship pairs,and each pair of relationships is labeled with sentiment values;finally,the percentage of evaluation words for evaluation attributes is counted to obtain entity and entity relationship data.(4)In terms of knowledge base construction,Neo4 j graph database is used for knowledge storage.Firstly,the review knowledge of the extracted laptop products is organized,and then,the review knowledge of the laptop products is stored using Neo4 j to complete the knowledge graph construction and serve as the knowledge base of the Q&A system.(5)An intelligent Q&A system based on the knowledge graph of product reviews in the e-commerce domain is designed,which consists of an interrogative intent recognition module and an answer feedback module.The question intent recognition module uses entity recognition and entity linking methods based on the BERT-Bi GRU-MHAT-CRF model for intent recognition of questions.The answer feedback module represents the question intent in the form of graph database query language and retrieves it in the knowledge graph to obtain fragmented answer information,which is organized and fed back to the questioning user.Finally,the question and answer system is implemented by combining all the above techniques,and after experimental testing,the developed system achieves the expected goal.