Transformer‐Enhanced Sentiment Analysis of Indian Farmers on Social Media Using Hybrid Deep Learning Models

ABSTRACT The COVID‐19 pandemic had a profound impact on Indian farmers, disrupting agricultural supply chains and exacerbating existing economic challenges. With a significant increase in social media usage among the farming community, educated farmers have increasingly turned to platforms like Twitter to voice their concerns. This user‐generated content offers researchers a valuable lens for understanding evolving farmer sentiments in real‐time and supporting responsive policy formulation. In this study, farmer sentiments were analyzed using a curated dataset of 40,000 tweets posted during recent periods of economic uncertainty and post‐pandemic recovery. Advanced machine learning and deep learning models, including CNN, LSTM, and hybrid transformers, were employed alongside diverse word embedding techniques such as Bag‐of‐Words (BoW), Term Frequency‐Inverse Document Frequency (TF‐IDF), Word2Vec, GloVe, BERT, and RoBERTa. All deep and hybrid models achieved strong classification performance, typically exceeding 85%. Notably, BERT–LSTM reached ∼93.0% accuracy (macro‐F1 ≈ 0.92) and RoBERTa–LSTM ∼93.5% (macro‐F1 ≈ 0.92), effectively capturing the nuanced emotional tone of the tweets. This study highlights the utility of modern NLP frameworks in analyzing real‐time sentiment data, offering researchers and policymakers a robust mechanism to monitor agricultural concerns and design timely, data‐driven interventions.