Global education has been revolutionized by Massive Open Online Course (MOOCs), which offer scalable access to high – quality education. Nonetheless, these platforms continue to face issues like low comparison rates, disengaged learners, and uneven satisfaction. In order to tackle these problems, this research suggests a sentiment analysis framework that blends contemporary transformer-based models with traditional Natural Learning Processing (NLP) methods. Specifically, DistilBERT, a lightweight transformer architecture designed for real – time applications, was used as a benchmark for VADER and multinominal Naïve Bayes with TF-IDF feature. Based on experimental result, Naïve Bayes performance decreased to 72.6% after class balancing, despite initially achieving an inflated accuracy of 92.6% due to dataset imbalance. A chatbot simulation that used sentiment predictions to inform adaptive, sympathetic responses was created to demonstrate the practical relevance. By enabling personalized feedback, lowering attrition, and facilitating adaptive learning interventions, the suggested framework lays the groundwork for incorporating sentiment – aware conversational agents into MOOCs. The responsible use of sentiment – driven chatbot in educational settings is highlighted by the ethical considerations that are covered, such as data bias, model interpretability, and learner privacy.