基于改进级联宽度学习的自适应认知诊断方法

An Adaptive Cognitive Diagnostic Method Based on Improved Cascade of Broad Learning System

针对现有的认知诊断模型信息利用不充分以及依赖局部作答信息而导致诊断精度低的问题,提出了基于改进级联宽度学习的自适应认知诊断方法。首先,提取题目的语义、参数等特征,采用无偏差加权进行融合。其次,提出了改进的级联宽度学习系统(improved cascade of broad learning system,ICBLS),旨在学习全序列作答信息,利用残差结构解决长序列学习遗忘的问题,采用网格搜索法确定最优参数组合,进而构建认知诊断模型。最后,经过非线性分类器实现知识状态的分类。以BP神经网络、Bi-LSTM、Bi-GRU为基线模型,在实际的接受性任务中进行了实验验证。结果表明,基于ICBLS的模型获得的最高模式准确率为95.74%,平均属性准确率为98.31%。并且,通过消融实验证明了题目的语义信息有利于模型更准确地发现被试的语言理解能力。

The existing cognitive diagnosis models could not use enough information and relied on local response information.Aiming at the problem of low diagnostic accuracy,an adaptive cognitive diagnostic method based on improved cascade of broad learning was proposed.Firstly,the semantic features and parameters of the items were extracted and integrated into vectors via an unbiased weighted method.Then,an improved cascade of broad learning system(ICBLS)was put forward to acquire the full sequence of the response information,and solve the problem of long sequence learning and forgetting with the residual structure.The grid search method was used to determine the optimal combination of parameters,and then a cognitive diagnosis model was built.Finally,the classification of the knowledge state was realized through the nonlinear classifier.With BP neural network,Bi-LSTM,Bi-GRU as the baseline models,experimental verification was carried out on the actual receptive task.The results showed that ICBLS model achieved the highest model accuracy of 95.74%and the average attribute accuracy rate of 98.31%.Moreover,the ablation experiment indicated that the semantic information of items could help the model to detect the language comprehension ability of the learners more accurately.