On detecting urgency in short crisis messages using minimal supervision and transfer learning

摘要: Humanitarian disasters have been on the rise in recent years due to effects of climate change and socio-political situations such as refugee crisis. Technology can be used best mobilize resources food water event a natural disaster, by semi-automatically flagging tweets short messages indicating an urgent need. The problem is challenging not just because sparseness data immediate aftermath but varying characteristics developing countries (making it difficult train one system) noise quirks social media. In this paper, we present robust, low-supervision media urgency system that adapts arbitrary crises leveraging both labeled unlabeled ensemble setting. also able adapt new where background corpus may available yet utilizing simple effective transfer learning methodology. Experimentally, our approaches are found outperform viable baselines with high significance myriad disaster datasets.