Generalizing to Unseen Disaster Events: A Causal View

2025-11-13

AI Summary

Social media disaster event classification suffers from event-related bias, impairing model generalization to unseen disaster types. To address this, we propose a causal learning-based debiasing framework that systematically models and disentangles event-specific biases from semantic intrinsic features—leveraging pre-trained language models, counterfactual reasoning, and bias disentanglement mechanisms to build a robust text classifier. Evaluated on three disaster classification tasks, our method significantly improves cross-event generalization, achieving up to a 1.9% F1-score gain over strong baselines. Our core contributions are: (1) formal identification and mathematical characterization of inter-event bias as a fundamental constraint on generalization; (2) a scalable causal debiasing architecture that enhances model adaptability to previously unseen disaster scenarios; and (3) a novel, interpretable, and robust paradigm for emergency information processing in social media.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Computer Vision: Bias, Fairness & Privacy

Application Category

• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Due to the rapid growth of social media platforms, these tools have become essential for monitoring information during ongoing disaster events. However, extracting valuable insights requires real-time processing of vast amounts of data. A major challenge in existing systems is their exposure to event-related biases, which negatively affects their ability to generalize to emerging events. While recent advancements in debiasing and causal learning offer promising solutions, they remain underexplored in the disaster event domain. In this work, we approach bias mitigation through a causal lens and propose a method to reduce event- and domain-related biases, enhancing generalization to future events. Our approach outperforms multiple baselines by up to +1.9% F1 and significantly improves a PLM-based classifier across three disaster classification tasks.

Problem

• Mitigating event-related biases in disaster monitoring systems
• Improving generalization to unseen disaster events using causal methods
• Addressing domain biases in social media disaster classification tasks

Innovation

• Causal lens approach for bias mitigation
• Reduces event and domain related biases
• Improves generalization to future disaster events