Research On Key Technology Of Question Answering System In Pension Insurance

Endowment insurance is a kind of universal welfare system for the citizens.It is the product of modern civilization and a symbol and sign of a democratic society.With the rapid development of socialism with Chinese characteristics,the endowment insurance system is also constantly improving.More and more people have been participating in endowment insurance.When people encounter various problems in participating in insurance,payment and receiving treatment,traditional search engines are insufficient to meet people's needs.Therefore,it is of far-reaching significance to study the key technologies of the question and answer system in the field of endowment insurance to meet the knowledge needs of the people who care for endowment insurance.This thesis starts from the above questions and studies the key technologies of the question and answer system in the field of endowment insurance.This thesis is divided into two parts,one is the research on domain ontology construction technology,and the other part is the research on classification model of domain problems.In terms of domain ontology construction,this thesis studied ontology construction techniques based on unstructured text corpora.In the field of problem classification,this thesis studied a problem classification model based on multi-features fusion.The construction of domain ontology can help people effectively organize and manage domain knowledge,realize the sharing and reuse of knowledge,and also help us to deepen the understanding of problems in the subsequent classification of problems.At the ontology construction stage,the concept of domain ontology is first extracted based on unstructured text.Then the domain ontology hierarchy and non-hierarchical relationships are extracted.The domain core ontology is constructed by combining the domain ontology core framework constructed by domain experts.The problem classification can not only reduce the choice space for the answer,but also can formulate different question and answer strategies for different types of questions,directly affecting the accuracy of the question and answer.In the problem classification stage,the corpus is first preprocessed,word vectors and part-of-speech features are constructed,and the above features are trained to obtain feature vectors and fuse the feature vectors with the ontology semantic features introduced in this thesis.Secondly,this thesis use the Bi-LSTM to includethe context information,the CNN to deeply mine the problem features,and the advantages of multi-convolution check feature to fully sample the features,and train the domain problem classification model to achieve Classification of the problem.Finally,this thesis carries out experiments on the above algorithms and models,and analyzes the experimental results.The improved hierarchical clustering algorithm based on word vector improves the effect of hierarchical relation extraction compared to the improved hierarchical clustering algorithm based on VSM.By introducing the ontology semantic features and the multi-feature fusion problem classification model compared to the single feature problem classification model,the accuracy of the problem classification model is improved.