语言测试成绩导出分数的算法优化与模型构建

The Algorithm Optimization and Model Construction of Derived Score in Language Testing

此研究以语言测试成绩的分数解释效力为研究对象,采用量化对比分析法,考察原始分数与导出分数在分数解释方面的优势与局限,以及导出分数的算法优化策略。研究发现,导出分数在分数解释、分数的可比性与可加性方面更具表现力;在原始分数偏度较大的情境下,基于log的Z分数比正态化的Z分数更具解释力;将原始分数转换为导出分数时,需要根据语言测试的功能和作用以及原始分数的实际分布样态选择相应的导出分数。分数转换模型有利于导出分数的实施与推广。

This paper studies the effect of score explanation in language testing.We use quantitative comparative analysis to examine the advantages and limitations of raw score and derived score,and the strategies for optimizing algorithms of derived score.Through the research,we find that derived score is more reasonable both in the score explanation,comparability and additivity of language testing,the log-based z-score are more explanatory than normalized z-score in the case of large deviation of raw score.When converting the raw score to a derived score,It is necessary to consider the function and effect of language testing,as well as the actual distribution of raw score.The score conversion model is conducive to the implementation and promotion of derived score.