基于深度置信神经网络的组卷难度预测

Test Difficulty Prediction Based on Deep Belief Networks

使用具有稳定性难度的试卷可以保证测试的公平性,在测试卷形成之前评估每一项试题的难度是非常重要的任务.提出一种基于深度置信网络的自动组卷新方法,该方法根据历史测试日志,设计基于深度置信神经网络框架来表示题干语义;接着应用注意力机制来度量句式的难度值,最后自动预测试题难度值.新方法在10万条测试日志上进行了实验,采用常见的均方根误差和教育心理学中常用的皮尔逊相关系数进行验证.实验结果表明,新模型在有效性和可解释性方面效果显著.

Using the test paper with stable difficulty guarantees the fairness so as to test the students＇ learning level. This paper presents a new method for automatic form test paper based on deep belief network. Based on the history test log, the new method is designed based on the deep belief neural network framework to represent the semantics of the question. Then the attention mechanism is applied to measure the difficulty value of the question, and finally the difficulty value is p^dicted automatically. The new method was tested on 100 000 programming questions, using the common root mean square error and the Pearson correlation coefficient commonly used in education psychology. Experimental results show that the new model is effective and interpretive.