Research On Difficulty Prediction Model Of Examination Questions Based On Text Extraction Of Association Information

With the intellectualization in the field of education,test resources in the education system and online learning platforms are increasing exponentially.How to avoid the waste of data resources,rationally and effectively use large amounts of education data to obtain personalized test resources,allocate appropriate topics for students with different knowledge levels and achieve personalized learning has become an urgent research topic.As a criterion to measure the rationality and fairness of a test,test difficulty mainly relies on expert markers in the traditional labeling of test difficulty,which has strong subjectivity and labor-intensive limitations.The thesis uses the abstract extraction method to extract the text information of the subject-related course resources,enrich the context semantics of the questions,and combine it with the difficulty prediction algorithm to improve the prediction effect.In the application scenario of multi-task feature test difficulty prediction,the main research contents are as follows:（a）Create an automatic abstract extraction model based on the text of title-related information.This paper summarizes the current research progress in the field of automatic extraction of text summaries based on topic-related information and difficulty prediction of neural network test questions,describes the ideas,principles,specific processes,advantages and disadvantages of automatic extraction of commonly used text summaries of related information,briefly introduces the methods of extracting text summaries of related information,and proposes a test difficulty prediction algorithm based on automatic extraction of text summaries of topic-related information.In the field of difficulty prediction,the concept of extracting text summary of related information is introduced.Firstly,a mining algorithm suitable for extracting text of related information of test questions is selected through comparison.Then,the main steps of constructing the text extraction model of related information are discussed in detail.Finally,the original data is preprocessed,and the text summary of related information is extracted by feature extraction and sentence scoring.（b）Construct a test difficulty prediction model based on text extraction of related information.Because the factors that affect the difficulty of the test are complex and difficult to extract,it is not possible to directly predict the difficulty.Therefore,this paper builds a test difficulty prediction model based on neural network,and optimizes the model with text extraction of related information.The model is implemented through in-depth learning using an end-to-end framework.The text summary of related information is extracted into the neural network model,and the insufficient data information resources are effectively solved by enriching the text content of related information of the test stem.The experimental results show that the difficulty prediction algorithm based on text extraction of related information has higher accuracy and feasibility than other difficulty prediction models.（c）Improve the difficulty prediction algorithm based on text extraction of related information.First,in the difficulty prediction module,Bi-LSTM is used instead of the recurrent neural network algorithm to compensate for the gradient disappearance caused by the RNN algorithm of the recurrent neural network.Then,when Bi-LSTM is used to obtain the local and sequence information of the test questions,the convolution neural network CNN is added to build a hybrid neural network model to improve the fitting ability of text-class data,to solve the problems that rely on lexical,grammatical and prior knowledge,and to make better text modeling.Finally,the attention mechanism is added to the hybrid neural network model to obtain the dominant information of the test with higher scores of sentences and to improve the model’s focus on the more relevant parts of the test in the semantically related sentences.The results of the experiment on the course "University Computer Foundation" show that the difficulty prediction algorithm based on text extraction of related information performs best in accuracy.