Study On Modeling Of Knowledge Base Question Answering Based On Deep Learning

Knowledge base question answering aims to extract the corresponding answers directly from the knowledge base by parsing the user's question sentence,which can greatly improve the user's search efficiency compared with traditional web search.The traditional knowledge base question answering method requires the design of a large number of rules and templates to transform the user's question into a structured representation,which is not suitable for large-scale knowledge bases.With its powerful feature learning and extraction capabilities,deep learning has achieved extremely dazzling results in the knowledge base question answering task.However,the existing deep learning-based knowledge base question answering still follows the steps of entity detection and relationship recognition,but such methods ignore the structural information contained in the knowledge base itself and the connection between these two tasks.This article is dedicated to researching the application of deep learning in knowledge base question answering.Aiming at the shortcomings and existing problems of existing knowledge base question answering algorithms,two knowledge base question answering methods based on deep learning and the design and implementation of knowledge base question answering systems are proposed.The specific research content is as follows:(1)Aiming at the traditional deep learning algorithm that does not consider the shortcomings of the structured information of the knowledge base itself,a question answering model based on the representation learning is proposed.In the model,the knowledge representation model is first used to map the entities and relationships in the knowledge base to a low-dimensional vector space,and then the question sentences are embedded in the same vector space through the neural network,and the entities in the question sentences are detected.In the vector space,the semantic similarity of knowledge base triples and question sentences is measured.Experimental results show that the algorithm can greatly improve the accuracy of the knowledge base question answering,and demonstrates the effect of knowledge base embedding on the knowledge base question answering task.(2)In view of the outstanding performance of the landmark BERT model in the field of natural language processing in the fields of reading comprehension and text matching,it is proposed to apply the BERT model to the knowledge base question answering.Firstly,it proves that BERT can significantly improve the various sub-tasks of the knowledge base question answering.In addition,combined with the technical route characteristics of the question answering task based on knowledge representation,the input of the BERT model is modified,and the sub-tasks of the knowledge base question answering are integrated into a model.And introduce the multi-task training strategy of loss function weighting and iterative training for training.The experimental results show that the final Q&A accuracy can be improved by combining training on the subtasks of the knowledge base question answering.(3)Different from the traditional web search system,the knowledge base question answering process and results are visualized,and the knowledge base question answering system is designed.First,import the knowledge base into the graph database,then use the front-end development framework Vue.js to design the web interface,and use Flask to encapsulate the interface for algorithms such as back-end language analysis and graph database calls.The above two knowledge base question answering methods are applied to the Simple Question public dataset to carry out experiments in this thesis,the effect of the model is measured through multiple sets of comparative experiments,and the experimental results are compared with other methods applied to the corresponding public datasets.The results show that the methods proposed in this thesis can greatly improve the performance of knowledge base question answering.In addition,the knowledge base question answering system designed in this thesis is also very easy to use and has good practicality.