Algorithm Design And Implementation Of Question Answering System Based On Knowledge Graph And Deep Learning

Task-based question answering systems are currently widely used in the field of intelligent customer service.For example,e-commerce assistants,after-sales return visits,telemarketing and other tasks can use question answering systems to replace humans in order to reduce labor costs.However,in practical applications,due to the vagueness of natural language itself,the accuracy of the information extraction part of many question answering systems is not very satisfactory,which directly reduces the quality of user experience of the question answering system.Moreover,most of the current question answering systems are constructed based on document retrieval,which is less flexible.Aiming at breaking through the above two limitations,this paper builds a task-based question answering system based on knowledge graphs and pre-training models,and focuses on the research and implementation of natural language information extraction and semantic matching algorithms.The main content of this article is as follows:(1)Aiming at solving the problem of poor flexibility of the document-based question answering system,this article uses a knowledge graph-based architecture to improve the freedom of knowledge storage and query in the question-and-answer system,and on this basis,the knowledge graph architecture is improved to make it better deal with questions that occur less frequently.(2)In view of the limitations of information extraction in the current question and answer system,this paper constructs a pipeline entity relationship extraction model that incorporates artificial features,which improves the accuracy of the pipeline extraction method and is more accurate than the joint extraction model,and has better robustness.(3)Aiming at the special form of the answer in this system,this paper designs an interactive text matching algorithm based on the pre-training model to sort the candidate answers,and on this basis,integrates entity relationship information to improve the answer sorting algorithm Accuracy.(4)Based on the above content,this paper builds a deep learning question and answer system based on knowledge graphs and pre-training models,and improves various functional modules.The system can respond flexibly to user questions under limited domain conditions,and is both accurate and robust,which reflects the practical application value of the above research.