Research And Implementation Of Automatic Question Answering System For Government Service

The administrative examination and approval items listed on the government service website are intricate,and people often have medical,housing,and education problems,and it takes a long time to find solutions to related problems.Because of the unfamiliarity with the process of handling affairs,the “difficult to ask things” and other issues reflected by the masses have become increasingly prominent.If can build an automatic question-and-answer system for government service,the masses can get the policy knowledge they want to ask in a short time.The construction of an automatic question-and-answer system for government service plays a key and important role for the masses to quickly understand the relevant government policy knowledge and administrative processes.In this paper,based on the actual needs reflected by the public,the research builds an automatic question and answer system for government service.The main work contents are:1.Propose a multi-feature question similarity algorithm.The algorithm combines four different question features,including question category features,question government entity word features,question sentence syntactic dependency features and question sentence semantic features,and builds a question classification model,government entity word recognition model and semantics model.The experimental results show that compared with Word2 Vec similarity algorithm and Term Frequency-Inverse Document Frequency(TF-IDF)similarity algorithm,the fusion multi-feature question similarity algorithm has higher accuracy.2.Suggest a method for recommending questions.First,build a user-based collaborative filtering problem recommendation model.Second,construct user portraits based on user behavior.Thirdly,a question keyword extraction algorithm based on textrank and inverse document frequency(Tk IDK)is proposed,and a question portrait is constructed based on the question keywords and subject words.Finally,based on the user portrait feature and the problem portrait feature,a question ranking model based on XGBoost+LR is constructed,and the recommendation results are reordered.3.Build a knowledge base of government service.The source data of the government service knowledge base consists of two parts.The first part comes from the questions and answers that the Shanxi government service network has left a lot of messages,and the second part comes from the administrative approval items announced by the Shanxi government service network.Sort the source data according to different departments to form a triple knowledge base containing questions,question types,and answers to questions,totaling 52508 items.The experimental results show that the F1 value of the automatic question answering system for government service is 0.812,the root mean square error(RMSE)of the recommendation model based on the collaborative filtering problem of users reaches 0.087461,and the AUC value of the problem ranking model based on XGBoost+LR reaches 0.919,which verified the feasibility of this method.