基于谱聚类的网上阅卷质量控制研究

2014-04-29 13:29:02贾志先
智能计算机与应用 2014年5期

摘 要: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。

关键词:网上阅卷;阅卷质量控制;谱聚类

中图分类号:TP301.6 文献标识码:A 文章编号:2095-2163(2014)05-

Research on Quality Control of Online Marking based on Spectral Clustering

JIA Zhixian

(Network and Experimental Teaching Center, Xinjiang University of Finance and Economics, Urumqi 830012, China)

Abstract: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 dimensions 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.

Key words:Online Marking; Quality Control of Online Marking; Spectral Clustering

0 引 言

主观题的评分依赖于评分员的主观判断,容易受到评分员的知识水平、综合能力、爱好、情绪、疲劳等主观因素的影响。不同评分员之间存在着主观差异,同一个评分员在不同的时间也具有主观不稳定性。这将导致评分员自身信度(intra-judge reliability)和评分员之间信度(inter-judge reliability)的降低,从而降低评分结果的信度[1]。

在评分的准确性(Accuracy/Inaccuracy)、严厉度(Harshness/Leniency)和集中度(Centrality/Extremism)等三个方面,评分员自身在多次评分时难以保持一致,不同评分员对于相同被试的评分也难以相同[1]。评分员在评分过程中存在的主要问题有:1)给分偏高;2)给分偏低;3)给分趋中,呈现“趋中评分”现象[2],即评分员较少给高分、低分,分数集中在中间段。在大多数考试的主观题评分中,普遍存在着这种现象。

在评卷过程中,有效地检测出评分员网上阅卷的质量问题,对于改进评分员的评卷水平,消除各个方面的因素作用于评分结果的影响,并且提高评分结果的信度,均有着难以估量的重要意义。……

登录APP查看全文