Research On Automatic Scoring Method For Subjective Questions

Automatic grading of subjective questions means automatically grading students’ answers to be assessed by establishing a grading model.Generally,it can be divided into automatic grading for short-answer questions with reference answers,and subjective questions for essays without reference answers.The question is automatically graded,and this thesis studies the former.With the development of the Internet and artificial intelligence technology,intelligence has emerged in all walks of life,and the education industry is no exception.More and more people have begun to pay attention to related research on intelligent scoring.Automatic scoring of subjective questions is an important part of intelligent scoring.link.At present,subjective questions are corrected manually,which consumes a lot of energy,and will lead to score errors due to the subjective factors of the corrector,and lose fairness.Therefore,it is of great significance to study the automatic scoring method of subjective questions.At present,the automatic scoring methods of subjective questions mainly include automatic scoring methods based on similarity and machine learning/deep learning.Learn the scoring rules.These two types of methods have important research value,but there are some problems: the similarity-based automatic scoring method for subjective questions considers single features,ignores the score point sentence matching between the answer to be evaluated and the reference answer;the machine learning/deep learning-based scoring method The automatic scoring method of subjective questions does not fully consider the interactive information characteristics of the answer to be evaluated and the reference answer,which all affect the scoring accuracy.In response to these problems,this thesis has completed the following research work:1.Aiming at the problem that the similarity-based automatic scoring method of subjective questions has a single similarity feature extraction and ignores the matching situation between the score points of the answer to be evaluated and the reference answer,an automatic scoring method of subjective questions based on similarity fusion is proposed.Keyword similarity and sentence semantic similarity calculate the score matching similarity,and then use the scoring formula to score.First,use the Text Rank algorithm to extract the keywords of each sentence in the answer to be evaluated and the reference answer,and calculate the keyword similarity between sentences according to the "Synonym Cilin" and related formulas;Sentences are matched,the semantic similarity between sentences is calculated,and the similarity of score points is calculated in combination with the similarity of keywords;finally,the score is calculated according to the scoring formula.Experimental results on related datasets demonstrate the effectiveness of the method.2.Traditional automatic scoring methods for subjective questions based on machine learning/deep learning often only consider the single text feature of the answer to be evaluated,ignoring the interaction information between the answer to be evaluated and the reference answer.For this reason,this thesis proposes a method based on BERT and mutual attention.An automatic scoring method for subjective questions based on the force mechanism.The BERT model uses the Transformer bidirectional encoder,which has excellent semantic feature extraction capabilities.In this thesis,the BERT model is used to extract the semantic features of the answer to be evaluated and the reference answer.On this basis,a mutual attention mechanism is introduced to extract the answer to be evaluated and the reference answer.Finally,the vector integrated with mutual attention is input to the prediction layer for scoring,and a data enhancement strategy is adopted for the dataset.Experiments on the dataset show that the method improves the accuracy of scoring,and it also verifies that the scoring effect can be effectively improved after data augmentation and expansion of the dataset.3.Collected the test data of a middle school and organized it into a Chinese subjective question automatic scoring data set to verify the validity and accuracy of the method proposed in this thesis.4.Design and implement an automatic scoring system for Chinese subjective questions.The system can automatically score the uploaded answers of the candidates to be evaluated,which proves that the method proposed in this thesis is feasible.