| In recent years,the scale of Chinese e-government market has been increasing y ear by year,The proportion of online government service users in the overall Interne t users has become more higher,people have more demand for government service.At present,many provinces have launched government service Q&A systems,but mo st of them use keyword matching technology to search information related to keywor ds in the database and return them to users,or build a FAQ database for governmen t affairs Q&A.The questions in the existing system are relatively fixed,so it’s hard to solve the practical problems of users.The government service Q&A system mainl y involves some service items,and its description has the following characteristics:1.A large number of affairs and exist internal association.Because of the differ ent administrative divisions,the total number of provincial,municipal,and county-lev el affairs is huge,and the types of affairs are also complicated.At the same time,a ffairs are not independent.There is also a case that one item is related to other item s,and the same item is related to different departments.2.The questions in the Q&A of government affairs are special.Due to the fact that some items have long names and juxtaposed meanings,the identification of gov ernment affairs in user’s questions is more complicated than the general named entity identification.3.The answer query is relatively complicated.Simple keyword matching cannot accurately locate the answer,use FAQ for Q&A.Users cannot customize the questio ns,and can only ask fixed FAQ.At the same time,due to the association between affairs and the office does not exist independently,the answer search may involve re asoning process.In view of the above problems,the main work of this paper are as follows:1)Construction of the government affairs knowledge base: In response to compl ex problems related to government affairs,the Selenium-based multi-thread crawler sy stem is used to capture Items and FAQ data,design the model layer of government affairs knowledge base by extracting concepts,attributes and relationships,and finally build a government knowledge graph based on the model layer.2)Question understanding in government affairs field: Aiming at the problem th at named entity recognition in the government affairs field is more complicated than general named entity recognition,this paper studies the methods of CRF,LSTM andsyntactic analysis.and uses the method of BERT-BLSTM-CRF to recognize the nam ed entity in the user’s question,the recognition accuracy reaches 92.23%.3)Answer query based on similarity and SPARQL: Aiming at relatively comple x question search,researched the similarity calculation method and SPARQL query m ethod,and constructed a common question answer search method based on TF-IDF si milarity calculation,and achieved 78 % Accuracy and guaranteed lower time consum ption.Constructed answer query of government knowledge base based on SPARQL,and uses natural language processing techniques such as word segmentation and part of speech tagging to query the government knowledge graph and return the answer t o users.The average accuracy rate was 73%.Finally,the Python was used to implement the automatic Q&A system in the go vernment service field,providing user-defined questions,hierarchical question answerin g,satisfaction evaluation and other functions.After evaluation,the automatic governm ent Q&A system designed in this paper has an accuracy of 77.3% for common quest ions in government service,66% accuracy rate for questions in government service,a nd the system’s response to user problem descriptions reached 91% accuracy rate. |
