A Question Answering System For Open Domain Is Researched And Implemented

The question answering task for open domain requires the question answering system to search through a knowledge document base,find the question-related paragraphs,and input the question and paragraph into the reading comprehension model to predict the question-answer.Unlike limited domain question answering,open domain question answering has no constraints on the content of the question query,with extensive knowledge topics.Therefore,predicting the correct answer highly depends on effectively retrieving text passages that match the question.However,existing research methods present an insufficient ability to clarify semantic ambiguity,which hinders the extraction of key features,resulting in poor paragraph retrieval and ultimately affecting the prediction accuracy.In this thesis,we proposed a retrieval model to solve the shortcomings of traditional retrieval models,with two major innovative models: a question clarification and a paragraph ranking model,presented as follows:(1)A question clarification model based on keyword prediction is proposed to solve the open-domain question answering,query content clarity,and feature ambiguity problem.The proposed question clarification model includes two main modules: keyword predictor and keyword selector.The keyword predictor aggregates the question semantic features through a deep convolutional network to determine the question query content.The keyword selector supplements the missing semantic information of the question by selecting the keyword information in the relevant paragraphs improves the clarification effect of the question and reduces its feature ambiguity.Experimental results show that the proposed model outperforms the baseline models on the question clarification task.(2)A paragraph ranking model based on bidirectional prediction is proposed,given that some questions do not match the retrieval results of the existing model paragraphs and the question query,the proposed model solves the aforementioned challenge by performing iterative clustering on the document paragraphs in the knowledge base,narrowing the retrieval scope,and retrieving the question-related paragraphs.Then,the answer is predicted according to the question and the relevant paragraph.The answer and the relevant paragraph represent the input of the question clarification model,and the clarification question is predicted in the reverse direction.Finally,apply to clarify questions to reorder the paragraphs in order to improve the matching of questions and paragraphs.The experimental results show that the accuracy of the proposed model in predicting the answers to ambiguous questions is higher than that of the baseline model.(3)A question answering system oriented to the open domain is implemented.To verify the practicability of the proposed model,combined with the practical application requirements of open domain question answering,a question answering system is designed and implemented to provide users with an open domain question answering service.