Design And Implementation Of Character Relation Question Answering System Based On Knowledge Graph

With the rapid development of Internet technology,the amount of data on the Internet is also growing rapidly.In the face of massive data,people require more concise and accurate information.The character relationship question answering system receives the user's natural language questions and returns to the user a concise and accurate answer,which effectively helps the user to quickly build a character impression.This thesis designs and implements a character relationship question answering system based on knowledge graph.The knowledge graph is used to store structured character information.The question and answer system uses the knowledge graph as a knowledge base to improve the efficiency and accuracy of question and answer.This system collects character information to build a knowledge graph of character relationships,and then uses question analysis and knowledge graph query to implement the question answering function.The main work of this thesis is as follows:(1)Based on the BiLSTM-CRF model,a BiLSTM-ATTE-CRF joint extraction model is designed and implemented for the task of knowledge extraction in the field of person relations.In order to reduce the accumulation error caused by tandem extraction,a joint extraction method based on joint annotation is adopted.In the BiLSTM-CRF model,a self-attention mechanism is added to improve the association between entity and relational feature words.Pre-trained Bert word embedding vectors and active learning methods are used to reduce the need for manually labeled data and improve the extraction effect.(2)Build the knowledge graph of character relations.Knowledge graph construction includes three submodules: data crawling and preprocessing,knowledge acquisition,and knowledge fusion.The data crawling and preprocessing submodule is to crawl character-related corpus and clean and label the corpus.The knowledge acquisition submodule uses the BiLSTMATTE-CRF model to implement joint extraction.The knowledge fusion submodule uses an alignment method based on entity similarity to integrate data from different sources and update the knowledge graph.(3)Design and implement the intelligent question answering system based on knowledge graph.The Q & A system includes an intelligent Q & A module and a web service module.The intelligent Q & A module implements the character relationship Q & A function and background management function.The intelligent Q & A module obtains user intentions and keywords through question analysis,and matches and knowledge reasoning in the knowledge graph based on the keywords.Design the concept graph structure and use the random depth traversal method to reason about the relationship of characters,mine the hidden information in the knowledge graph and improve the correct rate of question and answer.The background management function is used to maintain and update the question and answer system.Tests on the joint extraction model show that the active learning strategy can effectively reduce the cost of manual annotation.The BiLSTM-ATTE-CRF model designed in this thesis effectively improves the effect of joint extraction of entity relationships in the field of person relations.According to the function and performance test of the system,the system provides simple and effective Q & A and background management functions.The delay of intelligent question answering is within 1s,and the accuracy rate of question answering is more than 90%,and it has good practical ability.