【作 者】：
郁晓华， 战晓瑜
【关 键 词 】：
学习分析； 教师评语； 文本情感分析； 教师期望； 教师风格； 学习力
【栏 目】：
课程与教学
【中文摘要】：
随着教育评价改革的深化，教师评语作为重要的数据分析来源，虽带有丰富的情感色彩，但其非结构化的特点导致分析并不深入。采用细粒度的文本情感分析方法，基于教师属性和学生属性的情感识别，90237条教师评语文本背后的新价值被重新加以探析。研究构建了教师风格、教师期望和学生学习力三大情感词典，定义了六种教师风格和四个等级的教师期望，再设计情感公式计算了学生的学习力得分，并基于该得分将学生聚为六个类群。然后，结合学生的学习成绩探讨了不同教师风格、不同教师期望和不同学习力水平之间的师生匹配策略。研究发现相较于低学习力水平的学生而言，学习力水平高的学生更适合由温和且民主的教师指导，同时，教师评语中的低期望信息对他们学习产生的负面影响也相对较小。研究成果可引入相关教学支持系统，协助利用教师评语进行师生画像，还可为师生匹配决策提供支持。
【英文摘要】：
With the deepening of educational evaluation reform, teacher comments, as an important source of data analysis, are rich in sentiment, but their unstructured characteristics leads less in-depth analysis. Using a fine-grained text sentiment analysis method, based on the sentiment recognition of teacher attributes and student attributes, the new value behind 90237 teacher comments are re-explored. First, three major affective lexicons of teacher style, teacher expectation and student learning power are constructed, and six types of teacher styles and four levels of teacher expectations are defined. Next, the affective formula is designed to calculate students' learning power scores, and students are clustered into six clusters based on this score. Finally, the teacher-student matching strategies between different teacher styles, different teacher expectations, and different levels of learning power are explored in combination with students' academic achievements. It is found that students with high levels of learning power are more suitable to be guided by moderate and democratic teachers than those with low levels of learning power. At the same time, teacher comments that convey low-expectation messages have less negative impact on their learning. The results of the study can be introduced into relevant instructional support systems to assist in teacher-student profiling using teacher comments, and can also provide support for teacher-student matching decisions.