Research On Intellectual Evaluation Method Of Chinese Subjective Questions

Subjective questions are a type of question that often appear during examinations.Due to the variety and complexity of their answers,intelligent techniques for assessing subjective questions are not yet mature,and the evaluation of subjective questions is carried out by the viewer.However,through manual evaluation,the objectivity and fairness of the evaluation results cannot be fully ensured.Therefore,it is of some practical significance to conduct research on intelligent evaluation methods of Chinese subjective questions,so as to realize a better performance and applicable evaluation system.The main contents of this article in the study of intelligent evaluation methods of Chinese subjective questions are:First,we proposed a Chinese semantic similarity calculation model based on the dynamic semantic code in bidirectional LSTM.The model adopts encoder-decoder neural network frame structure.The encoder uses traditional bidirectional LSTM neural network,and the decoder uses bidirectional LSTM neural network with dynamic semantic code.The model takes the sentence-pair distributed word vector as an input and the output value as the similarity value of the sentence pair.By verifying with different content of data sets,the proposed model has good results in predicting Chinese sentence pair similarity.Next,Study the sentence pair generation strategy that generates sentence pairs of reference responses and student responses.The main method is to use natural language processing technology to pre-process short answer questions,reference answers and student answer to extract keywords and phrases.According to the keywords extracted from the test questions,determine the category of knowledge points examined by the test questions,and mark the score points in the reference answers.Calculate similarity from keywords and phrases extracted between sentences to get a sentence pair in a situation.Predict the remaining sentences of the reference solution and the student solution by the model.Get the pair of the sentence with the highest similarity among the student's solutions and generate a sentence pair based on the reference solution.Then,research on intelligent scoring of subjective questions.The scoring algorithm is based on the similarity value of sentence pairs,and each sentence pair corresponds to one knowledge point in the test question.In this paper,we linearly combine the values of the similarity of key information of sentence pairs,the similarity of main information ofsentences,and the similarity predicted by the model.Use particle swarm optimization to determine its parameters and form sentence pair similarity values.The student's answer score is obtained by multiplying the knowledge point score according to the sentense similar value.Finally,we constructed an "operation system" online test system and applied the Chinese subjective problem intelligence evaluation method in the exam evaluation function.The feasibility of the intelligent evaluation method has been verified through experiments and has good results.