Sentiment Analysis in Social Networks: A Case Study on Student Feedback
Manoj Kumar Srivastav, Somsubhra Gupta · Indian Journal of Science and Technology · 2026
Objectives: Social network applications such as Facebook, LinkedIn, Twitter, and WhatsApp are becoming an important part of daily communication. These platforms help users to share opinions, information, and feedback with others. Sentiment analysis of such content provides insights into users’ views, emotions, and engagement levels. This paper analyzes student feedback data collected from Kaggle to identify sentiment patterns in social networks. The study also aims to examine whether hybrid models can provide a balanced and interpretable alternative to deep learning models and transformer-based methods. Methods: This study uses several sentiment analysis approaches, including Support Vector Machine (SVM) along with TF-IDF features, Long Short-Term Memory (LSTM) networks, a hybrid TF-IDF–LSTM model, and transformer-based models like BERT and RoBERTa. The analysis is performed on a student feedback dataset consisting of 5,200 cases. The dataset is divided into training and testing sets using 80:20 ratios. Performance is measured using accuracy, precision, recall, F1-score, and a confusion matrix. The models are tested on student feedback data and compared under a unified experimental framework. Findings: In the LSTM model highest accuracy of 0.881 is found whereas in BERT and RoBERTa achieved accuracies of 0.880 each, while TF-IDF with SVM reached 0.865. LSTM and transformer-based models performs better than traditional methods in classification accuracy. In these results it is found that deep learning and transformer-based models give higher accuracy in sentiment classification. The hybrid TF-IDF and LSTM approach shows stable and competitive performance. It also needs less computational resources. So there can be a practical option for sentiment analysis of student feedback on social networks. Novelty: A hybrid sentiment analysis method combining TF-IDF and LSTM is presented. It combines frequency-driven statistical feature extraction and representation of sequential data for sentiment classification. The hybrid model is not designed to perform better than transformer models. The study tried to show hybrid models can be applicable in sentiment analysis of student feedback in social networks. Keywords: Sentiment Analysis, Social Networks, Student Feedback, LSTM, BERT, TF-IDF