Analyzing Middle School Students’ Distance Education Experiences in COVID-19 via Sentiment Analysis and Topic Modeling

This study investigated middle school students’ experiences with emergency remote education during the COVID-19 pandemic using natural language processing (NLP), sentiment analysis, and topic modeling techniques. A total of 2,739 valid responses from Turkish students (ages 9–15) were collected through open-ended survey questions regarding the perceived advantages and disadvantages of distance learning. Sentiment classification was performed using a semi-supervised machine learning approach, combining TF-IDF, Word2Vec, and FastText vectorization with five classification algorithms. The TF-IDF + support vector machines (SVM) combination yielded the highest performance (F1 = 0.85). Results show a total of 1,867 positive and 2,542 negative opinions, indicating that students generally adopted a more critical view of distance education. To explore the thematic structure of opinions, topic modeling was applied with six topics. Positive sentiments clustered around themes such as educational continuity, health protection, time savings, flexible scheduling, self-regulated learning, and digital literacy. Negative sentiments were dominated by themes including limited interaction, screen fatigue, perceived low quality, technical barriers, and structural inequalities. Findings suggest that while students appreciated the safety and flexibility of remote learning, they also faced significant pedagogical, physical, and technological challenges. The study contributes methodologically by demonstrating the effectiveness of AI-based text analysis and offers practical implications for designing more equitable and student-centered digital education models. These results underscore the importance of integrating NLP and machine learning tools into educational research to uncover deeper insights from student-generated content at scale.