Design And Implementation Of A Automatic Scoring System Based On Text Similarity For Subjective Questions

With the continuous advancement of artificial intelligence in the intelligentization of education,various test scoring systems emerge one after another,however,there are very few systems that can automatically score subj ective questions.At the same time,manual scoring for subj ective questions is often cumbersome,so,if the computer can be used to automate this work,the workload of the reviewer will be greatly reduced and the efficiency will be improved.Text similarity algorithm and its application in automatic scoring for subjective questions are the main function module of the system,the main improvements of algorithm are:1.On the basis of the cosine similarity between word-embeddings representing the semantic similarity between words,the word weight transition probability has a positive correlation with the semantic difference between words,so the average distribution of the word weight transition probability of the TextRank algorithm is improved,and the WE-TR algorithm is proposed and implemented to extract keywords.2.Combine WE-TR algorithm with WMD algorithm.When dealing with long texts,first use WE-TR to extract keywords,then use WMD to calculate text similarity,meanwhile,MWMD "inherits" the final word weight calculated by WE-TR instead of the normalized word frequency.The experimental results show that the effect of text similarity calculation of WE-TR-VMD is better than TF-IDF-VSM,LDA-JS and WMD.Besides,this paper also implements other function modules of the system,including the registration,login,and personal information management for use of teacher and student.For teacher's use,there are student management,question management,exam management,exam scoring,grade view,grade review,grade management,corpus management and update of the algorithm model for automatic scoring.There are also online answering and viewing answers and grades for student to use.Further,question management,exam management and corpus management all contain four sub-modules,which are creating or uploading,deleting,editing,and searching.Exam scoring implements two sub-modules,which are automatic scoring and manual scoring.Automatic scoring implements two methods which are text similarity algorithm and keyword matching method.Collecting student answers with high-quality that are positive samples to the corpus is implemented in grade management module,which enables the corpus having an automatic update mechanism,so that the automatic scoring effect of the text similarity algorithm can promote adaptively.The algorithm update model for automatic scoring is the unification of word vector training and update of text similarity algorithm model.The subjective questions of the system are not limited to fields or subjects,which support teacher-defined category of subject or knowledge point of questions and corpus,so the system have good versatility,scalability and certain practical significance.