Research On Automatic Evaluation Based On Submitted Historical Information

With the continuous development of code repository,software projects can be monitored through a collaborative platform,including each submission information and code differences.In this way,step-by-step code programming is conducive to students’ better understanding and completion of programming tasks,making students’ goals clearer and convenient to review in the future.However,current work is mainly focused on the study of code similarity.Student programs correction or feedback based on the submission history process has not been carried out,so the core of this paper mainly analyzes and deals with the students’ submission history process and tracks the students’ submission process coarse-grained,so as to achieve the purpose of effectively correcting students’ procedures.After statistical analysis of the submission history information of 72 student repository,it is found that the submission times,code modification times,submission commit and code modification fragments have positive effects on the evaluation of student repository scores.Firstly,information of submission commit and code modification fragment information are extracted according to submission history information,and then these two features are converted into vectors of the same dimension by word embedding to facilitate cosine similarity calculation.Then the Kuhn-Munkres algorithm is used to match the submitted commit and the code modification fragment respectively,so as to determine the submitted commit matching set and the code modification matching set.After that,the total similarity of the submitted commit and the total similarity of the code modification fragment of each student repository is obtained by averaging.Finally,the linear regression model is used to train the total similarity of submitted commit,total similarity of code modification,total number of submission and total number of code modification to determine the parameters in the evaluation model so as to obtain the most appropriate evaluation model.The experiment verifies that the research model proposed in this paper is excellent in matching the submitted commit and code modification fragment,especially in the accuracy rate,which lays a foundation for evaluating the student repository achievement model.After linear regression training,the automatic evaluation model in this paper can achieve the degree of almost close to the scores provided by teachers and can control the residual error of the least square method within a very small range.In addition,the data fitting effect of the model is also excellent,which successfully realizes the appeal of automatic evaluation.