基于聚类分析的锚测验等值样本选取方法研究

Research on Sample Selection Methods of Anchor Test Equating Based on Cluster Analysis

共同题等值方法受"锚题"选择、被试样本水平差距等多种误差因素的影响,在测试样本数据中存在一些噪声。以锚题的正确多少来推断试题的难易程度,会出现一些误差。利用聚类分析能发现数据集中自身隐藏的内蕴结构信息。应用k均值聚类算法,按照锚题模式向量对被试样本进行聚类,可以区分一些含有噪声的被试样本数据。以HSK高等考试的听力理解题为对象进行等值样本选取的研究,给出了被试样本的k均值聚类算法和共同题等值样本的选取算法。

Since the method of common items equating was affected by many error factors as ＂anchor items＂ choice and the gap be- tween the levels of the examinee samples, there were some noises in the sample data of the test. To infer the degree of difficulty of ex- amination papers by using the correct number of anchor items, there will be some errors. Using cluster analysis methods we can find its own hidden intrinsic structure information in the data set. By using k-means clustering algorithm, clustering of samples according to an- chor items pattern vectors can distinguish some test samples that contain noises. In this article, we choose listening comprehension ques- tions of the HSK advanced test as the research object that the equating sample selection, giving a k-means clustering algorithm of exami- nee samples and an equating sample selection algorithm of examination paper that based on common items.