Evaluation Model of Learner's Cognitive Level Based on RoBERTa Fused with CNN

The cognitive level is an essential indicator of the learning process evaluation. However, most of the existing cognitive level evaluation methods use manual coding or traditional machine learning methods, which cannot fully mine the implicit cognitive semantic information in unstructured text data. In this study, a fusion model named RoBERTa-CNN is proposed, which uses RoBERTa pre-training model to capture the cognitive semantic features implied between words in the sentence. Furthermore, it combines a convolution neural network to extract the global semantic features of sentences, to extract the deep-level features of the text. Using Bloom's cognitive goal classification theory as the framework of cognitive level evaluation, 9657 text data from MOOC were evaluated automatically. The results show that the mixed RoBERTa-CNN model has the best evaluation effect, and the overall accuracy rate of the six cognitive levels is 85.26%, demonstrating the model's validity.