Research On Subjective Question Scoring System Based On Similarity Combination

In the examination of schools and enterprises,Chinese subjective questions are indispensable.At present,the marking of Chinese subjective questions is still carried out manually.This method not only consumes a lot of time and workload,but also has different grading standards among different teachers.It is easy to be affected by the subjective factors of the examiners,which can not reflect the fairness of the examination.Therefore,the research and implementation of the automatic scoring system for subjective questions can improve the efficiency and fairness of teachers’ marking papers.Due to the high flexibility of Chinese subjective question text expression,sentence semantic information can not predict the content of the preceding and subsequent text,and the space of word vector representation is discrete,there is a certain gap between the subjective question scoring model based on text,semantic and keyword similarity,which makes it difficult to realize the automatic scoring of Chinese subjective questions This thesis studies the scoring system of subjective questions.First of all,keyword similarity ignores sentence information and text information,sentence semantic similarity easily ignores the relationship between sentence and the preceding and subsequent text;text similarity sometimes ignores the role of keywords.Therefore,in view of the lack of similarity in text,semantics and keywords,according to the accuracy of the overall semantics required in the answer of subjective questions,this thesis proposed a subjective question scoring model based on similarity combination,which could flexibly adjust the threshold value according to the scoring logic,which could improve the similarity of subjective questions.By judging whether the similarity value reached the threshold value in turn,if it did,the full score would be output.If all the similarity values did not reach the threshold value,the maximum similarity value was selected as the final similarity value of the examinee’s answer.Secondly,the subjective test questions were used as experimental data set,and the similarity combination method was used to score,and the variance and deviation rate were used as the measurement standards to verify the scoring model based on similarity combination.The scoring results based on similarity combination were compared with the scoring results based on word2 vec,doc2vec,TF＿IDF and the automatic scoring of subjective questions in the learning pass system,the experiment showed that the variance and deviation rate of the scoring results proposed in this thesis were small and stable.Then,using the object-oriented development method,UML model is used to analyze the requirements,structure design,activity,sequence,class design and database design of the subjective question scoring system.Finally,python language,Django framework and pychar were used as the integrated environment for development.A subjective item scoring system based on similarity combination was realized and tested.The test results showed that the prototype technology route of the subjective question scoring system based on similarity combination is feasible,which laid a certain foundation for the practicability of the system.