基于谱聚类的网上阅卷质量控制研究

Research on Quality Control of Online Marking based on Spectral Clustering

在网上评分过程中,存在着评分员给分偏高、偏低和趋中等问题。对于评分员的评分结果,可从用平均分、标准差、离差、阅卷量、问题卷等维度进行评价。给出了评分员的评分测量对象、距离和相似度的定义。结合评分员的评分结果度量维度参数,利用谱聚类算法可以有效地检测出评分员在主观评分过程中存在的一些问题。以维吾尔语水平测试中口试的评分结果,进行了网上阅卷质量检测的研究。实验表明,检测方法有效。

In the online marking process, there are some problems that the scores given by exam markers may be too high, too low, or too trending to the central, overall. For these marking results, the researchers can evaluate it from some dimen- sions as average, standard deviation, deviation, quantity of marking and quantity of error marking. This article has given the definition of the marking measurement object, the distance and the similarity of that. Using the spectral clustering algorithm, the paper can effectively estimate some problems existing in marking of subjective. According to the marking results of oral examination in Uyghur proficiency test, the paper also has done some research on quality control of online marking. The experimental results show that the method is effective.