Finding Evidence of Critical Thinking in E-Learning: An Application from Sentiment and Discourse Analysis Perspective

Critical thinking is arguably one of the most cherished, albeit enigmatic, of the cognitive skills in digital learning. Although significant behavioral data are obtained by e-learning systems, few e-learning systems are able to detect evaluative or reflective thinking in a meaningful way. This study introduces a novel machine learning framework for the identification of critical thinkers in which behavioral contradictions are modeled - e.g., learners who demonstrate high levels of content engagement but assign low ratings to the material. Leveraging multilingual datasets in both English and French, we combine implicit (watch percentage) and explicit feedback (user ratings) to devise a binary classification model using Random Forest with class balancing. We eschew conventional proxies like test scores and instead consider cognitive dissonance as a cue about how evaluative thinking has been undertaken. The model is promising for identifying students whose data show intellectual engagement and skepticism. The results address a research need and make an important contribution to the literature by providing a scalable theory-based, and cross-culturally adaptable methodology for identifying students’ critical thinking activity and in turn increasing the pedagogical relevance of IA.