Research And Implementation Of Natural Question Analysis And SPARQL Query Generation Method

As the amount of structured knowledge increases over time,and the knowledge base in the format of RDF is widely applied in different fields,bringing about the situation that the demand of end users to access this knowledge increases over time.Especially the question answering system is being of greater importance to end users,compared with the traditional search engine.Via adopting natural language,question answering system provides direct access to data,and keeps the end users free from process,such as data modeling,vocabulary construction,and query language technology.The interaction between the end users and the system is only about input question.Finally,obtaining the precise and concise answer is much more convenient than accessing to Web page for the answer.Nevertheless,the question answering system that adopts knowledge base as the underlying data storage cannot access the data only by keyword matching,but through structured query language.Due to the limitation that processing the query language has to be professional,parsing natural language and generating structured query language has become a research focus in the question answering system oriented to knowledge base.This thesis proposes deep question answering and SPARQL query generation method,based on natural language question for semantic analysis.In combination with the question analysis techniques,including syntactic structure analysis and dependency parsing analysis,the semantic relationship between words in the question is obtained.Semantic query graph can mark the topic words,identify the entity words and category words of the question,extract the semantic units in the question,and construct the structure of relational triples.In order to map natural language to knowledge base resources and improve the accuracy of entity linking,this thesis adopts the entity link method based on knowledge base,adopting Wikipedia to construct external entity mention dictionary and relational mention dictionary,and finally combines the knowledge base to match entity and predicate path,so as to optimize the semantic query graph.In order to match the SPARQL query mode corresponding to the question,the method of question classification based on Bi-LSTM is adopted to obtain the category of the question,and the correct query mode is selected in the SPARQL query generation stage.Due to the low accuracy of complex questions in the question and answer system at present,via adopting question decomposition technology to decompose the questions that are complex and decomposable,rewriting sub-questions,and analyzing sub-questions,this thesis simplifies the process of complex questions so as to enhance the accuracy of complex questions.Finally,according to the above solution,this thesis designs and implements a question and answer system for knowledge base,realizing the problem analysis functions,such as,semantic analysis,question classification and question decomposition,so as to obtain the intent of end users to ask questions.Then,based on the structure of the dependency analysis,the semantic query graph model is defined and the conversion from the natural question to the semantic query graph is realized.After that,via traversing the semantic query graph and based on SPARQL query definition,the SPARQL query generation based on the semantic query graph is implemented,and then via retrieving the generated SPARQL query statement,the SPARQL query engine returns the answer.Finally,through a visual interface,the interaction with the end users is completed.The question and answer system reaches an F value of 83% on the QALD Web Question dataset,compared with Ask How,ONIL Querio Da LI,and Xser,the method proposed in this thesis makes a significant improvement on performance in terms of the natural language query generation methods.