Research And Implementation Of Thangka Character Question Answering System Based On Knowledge Graph

Thangka is a kind of scroll painting,which is an important material historical material to explore the development of China’s history and culture,economy and society,national integration and so on.This article studies and implements the Thangka character question and answer system based on knowledge graph.This article includes the following parts:Firstly,the knowledge graph of Thangka characters is constructed to provide a knowledge base for the question and answer system.By means of web crawlers and rules,the relevant data of Thangka characters are obtained from websites and e-books,and the data are cleaned and sorted.Finally,the triples are stored in the neo4 j map database to realize the construction of Thangka character knowledge graph.Finally,the visual tools provided by neo4 j are used for query and display.Secondly,a sentence error correction method in Thangka domain is proposed.The n-gram language model based on domain error prone word set and based on word and word granularity is used to detect the errors of sentences.The sound / shape near words and editing distance are used to form the wrong word candidate set.The wrong word candidate set is brought into the original sentence.After calculating the confusion degree of sentences,the corrected sentences are sorted.Thirdly,based on the classification algorithm,the problem intention is deeply analyzed and the Thangka problem classification is realized.In the stage of question intention analysis,mark the training data of question type,divide the Thangka question into 15 categories,sort out 201 Thangka domain question and answer questions,and then train the marked data.KNN,naive Bayes and decision tree algorithms are compared.Finally,the classification of input questions is realized based on Naive Bayes classifier model.Fourthly,Tangka character entity recognition based on Bi LSTM-CRF.In this article,the CRF layer is added to the output layer of the two-way LSTM model to restrict the labels in the context,then the entities in the question are extracted and filled into the cypher template,the natural language question is transformed into the knowledge map query language,the relevant entities contained in the question are mapped into the Thangka character knowledge graph,and the knowledge graph is retrieved,Finally,the result is returned to the user.Finally,the retrieval of Thangka characters and the visual display of character relations are realized by using Echarts.The design and implementation of Web question and answer platform are completed based on flask framework.The Thangka character question and answer system based on knowledge graph is verified from theory to method.