Research On Misunderstanding Detection And Repair Of Short Text Question Answers Based On BERT

With the rapid development of Internet technology,online learning has become the choice of more and more students.At present,there are a large number of online learning platforms at home and abroad,such as Tencent Classroom and MOOC in China,OpenStax Tutor and Coursera in foreign countries.These platforms allow students to have more convenient access to a variety of knowledge.At the same time,it is also conducive to the teaching work of teachers.Learning analysis[1]refers to recommending an appropriate learning program to different learners by studying their learning behaviors and habits.In order to better intervene learners,high quality behavioral data of learners are needed.How to obtain these data is a very worthy of research,among which the problem of misunderstanding detection and repair for short text questions is an urgent one to be solved.These essay answered this question misunderstanding can reduce the students’ learning efficiency,but also reduces the teacher’s teaching efficiency,is not conducive to the teacher on relevant teaching work,and make learning the results of the analysis with large deviation between actual situation,so the detection and repair this essay question answer misunderstanding is a very meaningful work.Based on online learning community that may occur in the essay answer the misunderstanding of the problem,put forward the following two kinds of online learning community essay answer the misunderstanding of detection and repair of this framework,the two framework defines two essay answer the misunderstanding of detection and repair of the system structure,the improvement of the second frame is the first frame,Both frameworks can independently and automatically detect and repair misunderstandings in the online learning community.The main work of this paper is as follows:Firstly,this paper proposes a Bert-based framework for misunderstanding detection and remediation of short text questions.The framework consists of three steps.First,the vector matrix of Bert input layer is obtained by three embedding methods.Secondly,the vector matrix obtained by preprocessing is input into the Bert pre-training model.Thirdly,Softmax is used for normalization processing and end-to-end adjustment of the model to make it fit into a misunderstanding detection model that can identify whether there are misunderstandings in short texts.Finally,the correct answer is used to mark the misunderstanding and push to repair the misunderstanding,that is,if the model finds a misunderstood answer on the data set,it will find the problem number based on the answer number,and find the correct answer based on the problem number,and mark the misunderstood answer with the correct answer.Both the question and the correct answer are pushed to the respondent.Then,this paper proposes a framework for misunderstanding detection and repair based on BERT and LSTM in short text question answering.The framework is divided into three steps.Firstly,the short text is transformed into a word embedding vector with full contextual information extracted by the Bert pre-training model.Secondly,the obtained vector is input into the LSTM model for training,and a misconception detection model of short text question is obtained.Finally,correct answers are used to mark the misunderstanding and push the method to repair the misunderstanding.Finally,this article through OpenStax real online learning community on the Tutor data set,and the proposed method and contrast method proposed by other researchers,the experimental results show that the proposed two methods to a certain extent,can detect and repair the essay answered this question that exist in the online learning community of misunderstanding,It has certain guiding significance to study analysis and research.