A HYBRID TRANSFORMER BASED MODEL FOR DISASTER TWEET CLASSIFICATION USING BERT AND RoBERTa

In the era of social media, platforms such as Twitter play an essential role in real-time disaster reporting, offering immediate access to firsthand information during emergencies. This research presents a novel hybrid deep learning model for classifying disaster related tweets by integrating two state-of-the-art transformer architectures: BERT and RoBERTa. Our approach leverages the complementary strengths of each model by independently encoding the same tweet using both architectures, and then fusing their mean pooled representations to generate a more robust feature set for final classification. This dual stream method captures rich semantic nuances and contextual variations, substantially improving performance on noisy, unstructured text data. Extensive experiments were conducted on a carefully curated dataset of disaster and no disaster tweets. The experimental results demonstrate that our hybrid model significantly outperforms individual transformer models in terms of accuracy, precision, recall, and overall robustness. Detailed analysis of the combined outputs reveals that the hybrid approach effectively mitigates model specific limitations and enhances semantic representation. This work provides valuable insights into Mult transformer fusion strategies and suggests that integrating diverse pretrained models can yield substantial improvements in natural language processing tasks for real world crisis management applications. Furthermore, these findings exhibit strong promise.