基于联结主义的连续记分IRT模型的项目参数和被试能力估计

运用联结主义中的级连相关模型对于小样本条件下的连续记分项目反应理论 (IRT)模型的项目参数和被试能力进行了估计。一组被试对于一组项目的反应矩阵作为级连相关模型的输入 ,这组被试的能力θ或该组项目的参数a、b和c作为该模型的输出 ,对神经网络进行训练使之具备了估计θ,a ,b或c的能力。计算机模拟的实验表明 ,如果测验中有少量项目取自于题库 ,就可以运用联结主义方法对IRT参数和被试能力进行较好的估计

ITEM PARAMETER AND PERSON ABILITY ESTIMATE FOR CONTINUOUS IRT MODEL WITH CONNECTIONISM

The item parameter and person ability in continuous scoring Item Response Theory (IRT) model of small sample were estimated with cascade correlation model of connectionism. A new method to estimate the item parameter and person ability was suggested which was based on artificial neural network (connectionism),it was totally different from the traditional method which was based on statistics. The response matrix of a group of subjects to a set of items was used as the imput of cascade correlation model (neural network),the ability θof this group subjects or parameter a?b or c of this set of items was used as the output of this model. A set of neural networks were trained by them. After training, the neural networks were able to estimate the subject ability θand item parameter a,b or c. If a new response matrix was input, the neural network could output the estimate value of θ,a,b or c. The errors between estimate values and true values were very small. The results showed: If there were a few items from the item bank, the IRT parameter of continuous scoring item and person ability could be estimated under small sample condition.