【摘要】 在工程教学课程评价中，由于存在不可观测的混杂因子，使教师在进行课程目标达成情况评价时无法得到真正可靠的数据，故而影响后续的教学持续改进工作。针对此问题，本文提出一种结合神经网络认知测量学和因果推断科学的去混杂方法。首先，根据学生平时的练习数据构建神经网络认知诊断模型，得到学生对知识点的掌握程度并作为学生能力的衡量指标；然后将学生的能力评估结果作为该案例因果推断模型中的中介变量数据；最后通过前门调整法得到去混杂后的实际课程教学对课程目标达成情况评价的因果效应。本文以湘潭大学某学年电子信息类专业本科生专业课“电源技术”为案例，对该学年的课程目标达成情况评价结果进行修正，得到该学年实际的平均课程目标达成情况评价结果为88.92%。结果表明，该方法可以有效屏蔽混杂数据，帮助教师在实际教学中进行更加可靠、公平的课程目标达成情况评价。更多还原

【Abstract】 Due to unobtrusive confounding factors in the evaluation of engineering teaching curriculum, teachers cannot get reliable data when calculating the evaluation of the achievement of curriculum objectives, thus affecting the teachers to carry out the continuous improvement of teaching in the future. To solve this problem, this paper proposes a de-confounding method combining neural network cognitive measurement and causal inference science. Firstly, the neural network cognitive diagnosis model is constructed according to the students’ daily practice data, and the students’ mastery of knowledge points is taken as the measurement index of students’ ability. Then, the students’ ability assessment results are used as the mediating variable data in the causal inference model of this case.Finally, through the method of front door adjustment, the causality effect of actual course teaching on the evaluation of the achievement of the course goal is got. Taking the professional course Power Supply Technology for undergraduates majoring in electronic information in Xiangtan University in an academic year as a case, this paper revises the evaluation of the achievement of course objectives in that academic year, and obtains the actual average achievement evaluation of course objectives in that academic year as 88.92%. The results show that this method can effectively shield confounder data and help teachers get more reliable and fair evaluation of the achievement of curriculum objectives in actual teaching.更多还原