Research Of Question Answering System Based On Knowledge Base For Government Affairs

The development of Internet has promoted the reform of government services.At this stage,China is vigorously promoting the "Internet Plus government affairs" action plan to build online intelligent service systems such as "one website for all affairs,one window handles all affairs,multiple departments jointly handle one affair,one-click Intelligent Question Answering".However,the poor quality of the response content and low accuracy rate of the existing government affairs question-answering(QA)service make the performance of them not good.In order to improve the service level of government affairs QA,this paper studies the construction of question answering over knowledge base(KBQA)for government affairs.The main work includes:(1)Construct government affairs knowledge base to provide data support for government QA service.Firstly,this paper sorts out the knowledge system and construct the domain ontology of the government affairs field under the guidance of domain experts,and then constructs the knowledge base in a top-down way,thus ensuring the authority of knowledge and solving the problem of low quality of reply content.(2)Propose a government entity linking model based on pseudo-Siamese network to solve the task of topic entity selection.the model decouples the feature extraction process of questions and candidate entities by the pseudo-Siamese network,which effectively reduces the computational complexity of the model.At the same time,the model introduces the context information of the entity in the knowledge base to enhance the semantic characteristics of the entity,which makes the model have better distinguishing power on similar entities,and thus improves the accuracy of the entity linking task.(3)Propose a question solving model based on bidirectional attention mechanism to solve the task of candidate answer selection.Firstly,the model uses the question semantic information as attention,strengthens the semantic vector representation of candidate answer,then uses the feature information of the candidate answers as attention,strengthens the semantic vector of question,and finally selects the results sorted by the cosine distance.The introduction of bidirectional attention mechanism is helpful for the proposed model to fully capture the semantic interaction information of questions and candidate answers,which makes the generated semantic vector representation more accurate,thus improving the accuracy of question and answer task.(4)Design and implement the government affairs KBQA to provide users with accurate and efficient government affairs QA service.The government affairs KBQA includes knowledge management module and QA service application.The knowledge management module realizes a multi-user-oriented multi-level knowledge management function of administrative divisions,which is convenient for administrators to maintain government affairs knowledge within the scope of authority.