Research And Implementation Of Automatic Question-answering System Based On Attribute Graph In Financial Field

In the era of big data,new requirements have arisen for the efficiency and accuracy of information retrieval.As a kind of advanced information retrieval method,question-answering system has received extensive attention.Knowledge graph stores various kinds of knowledge through a structured representation form,which effectively improves the accuracy of information retrieval and becomes an important data support method for automatic question answering systems.However,there are still some problems in the automatic question answering system based on knowledge graph.First,it focuses more on open fields and lacks in-depth research in vertical domain.In the financial field,the length of the entity name is long,and problems usually contain complex constraints.Existing question answering algorithms are not well matched.Second,it lacks the ability to understand the semantics of complex natural language problems.Current obstacles include poor relational reasoning ability,ambiguous entity recognition boundaries,and difficulty in identifying implicit conditional constraints.In observation of the drawback above,the thesis conducts the following research works as follows.(1)Research and implement an automatic question-answering method based on attribute graph through principle analysis and model improvement of existing automatic question answering model.In this method,natural language question is understood by means of semantic parsing,and then it is converted into logical query language.Finally,the answer to the question is queried in the knowledge graph.The property attribute is easy to store the relevant knowledge in the financial field,and its query language is very suitable for representing complex problems with constraints.Experiments show that the proposed method achieves promising performance effectively improves the ability of automatic question answering methods in answering complex questions in the financial field.(2)Research and implement the implicit mention recognition algorithm based on joint semantic embedding.In order to further enhance the relational reasoning ability of the automatic question answering model and to accurately identify some constraints implied in the question.With the automatic question answering method,this thesis proposes an implicit mention recognition algorithm based on joint semantic embedding.The method combines question with graph,realizing the inference of implicit constrained questions in the financial field,which further enhances the accuracy of question answering.(3)Design and implement an automatic question answering platform in the financial field.In this thesis,an attribute knowledge graph is built by collecting relevant knowledge in the financial field on the Internet.Taking it as the data base,the proposed automatic question answering method is regarded as the core.Combined with the industry demand analysis in the financial field,an automatic question answering platform system in the financial field was successfully designed and built.The system is capable of answering natural language questions from users.At the same time,it can provide users with efficient information retrieval and graph visualization service.It has passed the system test.