A Machine Learning-Based Sentiment Analysis of Article 370 Tweets to Support Government Policy Decisions

This study proposes a robust sentiment analysis framework to evaluate public opinion on the abrogation of Article 370 using Twitter data. The methodology begins with tweet collection through the Twitter API, followed by systematic preprocessing. Sentiment labels were generated using the Text Blob lexicon-based polarity scoring approach to facilitate the construction of a large-scale sentiment dataset. Features are extracted, and the dataset is split into training (80%) and testing (20%) sets. A variety of models—including lexicon-based approaches, traditional machine learning algorithms, and ensemble learning techniques are trained and optimized using hyperparameter tuning. Additionally, a hybrid CNN–LSTM deep learning model is employed to capture both spatial and temporal dependencies in the text data. Experimental results reveal that the tuned Voting Ensemble model achieved the highest agreement with lexicon-derived labels, achieving an accuracy of 94.05% and an F1-score of 95.7%. The CNN–LSTM model also demonstrated strong performance. Lexicon-based polarity trends show that the dataset we looked at had mostly positive feelings during the time we chose. The results show how well ensemble and deep learning methods work together for automated sentiment classification. Future work could include validation datasets that have been annotated by people, support for multiple languages, more advanced transformer-based models for detecting sarcasm and emotion, and analysis of temporal sentiment trends.