Hybrid Deep Learning-Based Framework for Students’ Cognitive Analysis in Online Learning

This paper presents a framework for categorizing students’ discussion posts into different cognitive levels to gain insights into student engagement and enhance their online learning experience. The proposed approach investigates various hybrid deep learning models, which include CNN-BiLSTM, ANN-BiLSTM, and BERT-BiLSTM, within an educational framework called Interactive, Constructive, Active, and Passive (ICAP). These models classify students’ posts into interactive, constructive, active, and passive categories, which also quantify the cognitive engagement levels from highest to lowest accordingly. We analyzed textual attributes such as sentiment polarity, subjectivity, n-grams, and named entity recognition within the above hybrid models to extract deeper insights into the cognitive content of the forum posts. These features are also further refined using pre-trained BERT embeddings to enhance the classification accuracy. In addition to cognitive analysis, we employ Latent Dirichlet Allocation (LDA) based visualization to uncover the underlying themes and topics within the discussion posts. By using hybrid deep learning models for students’ cognitive analysis and LDA-based visualization of discussion topics, this study demonstrates the effectiveness of the proposed framework in improving students’ engagement in online learning.