Context-Aware Transfer Learning Approach to Detect Informative Social Media Content for Disaster Management

In the wake of disasters, timely access to accurate information about on-the-ground situation is crucial for effective disaster response. In this regard, social media (SM) like Twitter have emerged as an invaluable source of real-time user-generated data during such events. However, accurately detecting informative content from large amounts of unstructured user-generated data under such time-sensitive circumstances remains a challenging task. Existing methods predominantly rely on non-contextual language models, which fail to accurately capture the intricate context and linguistic nuances within the disaster-related tweets. While some recent studies have explored context-aware methods, they are based on computationally demanding transformer architectures. To strike a balance between effectiveness and computational efficiency, this study introduces a new context-aware transfer learning approach based on DistilBERT for the accurate detection of disaster related informative content on SM. Our novel approach integrates DistilBERT with a Feed Forward Neural Network (FFNN) and involves multistage finetuning of the model on balanced benchmark real-world disaster datasets. The integration of DistilBERT with an FFNN provides a simple and computationally efficient architecture, while the multistage finetuning facilitates a deeper adaptation of the model to the disaster domain, resulting in improved performance. Our proposed model delivers significant improvements compared to the state-of-the-art (SOTA) methods. This suggests that our model not only addresses the computational challenges but also enhances the contextual understanding, making it a promising advancement for accurate and efficient disaster-related informative content detection on SM platforms.

@article{Saleem2024,
  title     = {Context-Aware Transfer Learning Approach to Detect Informative Social Media Content for Disaster Management},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {1},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Saima Saleem and Monica Mehrotra},
  doi       = {10.14569/IJACSA.2024.0150167},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150167}
}