ASPECT BASED SENTIMENT ANALYSIS EDUCATION FEEDBACK FOR IMPROVING EDUCATION QUALITY AND PERFORMANCE USING MACHINE LEARNING AND DEEP LEARNING

Abstract

The education domain generates a substantial volume of textual feedback through student reviews, online learning platforms and course evaluation, providing valuable information for evaluating the teaching quality, course material, course design and institutional services.  The Aspect based sentiment analysis (ABSA) has gained noteworthy attention as useful technique for getting fine grained options regarding the education domain. In this research study, we have implemented five machine learning methods such as Support vector Machine (SVM), Random Forest (RF), and Logistic Regression (LR), Naïve Base and XGBoost and two deep learning architectures (LSTM and BiLSTM).  The models are trained and evaluated on 13,070 annotated student review instances spanning four polarity classes. Due to data diversity, experimental results indicate that XGBoost achieved the highest accuracy (87%) among the seven models capturing contextual and semantic relationships with educational texts.  Misclassification analysis reveals that all models struggle most with neutral polarity due to severe class imbalance.  With BiLSTM showing the most balanced tradeoff between majority class accuracy and minority class sensitivity, these findings provide actionable insights for education data mining systems aiming to automate the analysis of student feedback at scale.