Research On Common Sense Question Answering Model Based On Knowledge Grap

Common sense question answering is a research topic witch has great application value in the field of natural language processing.It aims to allow machines to simulate human thinking and give correct answers to the questions asked.However,since the problem formulation often does not contain background knowledge,it poses a huge challenge to traditional machine learning methods.With the increasing refinement of deep learning technology and the continuous output of research results,common sense quizzes have been widely used in industrial and commercial fields,and at the same time promoted the progress of interdisciplinary subjects such as bioinformatics,smart medical care,and social sciences.Research and social application value.Although the traditional question-answering model that implements answer reasoning based on articles or paragraphs has surpassed humans in terms of accuracy.However,the question answering task that relies on common sense knowledge is still an unsolved problem in the field of natural language processing,and it is also one of the scale to evaluate whether the machine really has reasoning ability.In order to make the prediction effect of the question answering model better and ensure the accuracy and reliability of the predicted answers,this paper conducts research on the common sense question answering task based on the knowledge graph according to the characteristics of the common sense question answering task.The main research content is as follows:(1)Aiming at the multi-background knowledge fusion problem,a context-aware and knowledge-augmented common sense question answering model CAARK is proposed.First,random walks are used to obtain the two-hop related entities of the answer entity in the commonsense knowledge map Concept Net,and then the given question and answer text is fused to generate enhanced questions and input to the pretraining model Ro BERTa.Finally,context-aware attention is used to strengthen the relationship between questions and answers.Semantic representation between.Experimental results show that after effectively introducing external knowledge,the CAARK model performs well on the Commonsense QA 1.0 dataset,providing a new paradigm for solving common sense QA problems.(2)Aiming at the sparsity and noise interference of knowledge graphs,a common sense question answering model CAARK-AGN is proposed that integrates pre-trained text features and statistical features.The model focuses on the original word frequency and word distribution information,and uses TF-IDF value as an additional feature to selectively enhance external knowledge features.The matching entity definition in Wiktionary is used together with the knowledge map Concept Net as external knowledge to assist the model to choose the correct answer.In the answer reasoning process,the attention mechanism is applied to deeply fuse the original semantic representation vector and the output vector of additional statistical information,and jointly reason the correct option.The experimental results show that the accuracy of the CAARK-AGN model on the Commonsense QA1.0 dataset has been further improved.In summary,the two common sense question answering models based on knowledge graphs proposed in this paper alleviate the disadvantages of traditional machine question answering models lacking background knowledge to a certain extent,and can effectively deal with the noise interference caused by increasing knowledge.The experimental results verify the effectiveness of the model.The key technologies of common sense question answering involved in this paper can be applied to scenarios such as intelligent customer service and chatting robots,and are also the basis for multimodal tasks such as visual question answering,which have certain theoretical significance and application value.