CReMa: Crisis Response through Computational Identification and Matching of Cross-Lingual Requests and Offers Shared on Social Media

TL;DR

CReMa tackles the challenge of automatically identifying and matching crisis-related requests and offers across multilingual social media. It fuses textual embeddings from crisis-specific, cross-lingual encoders with non-linear temporal and spatial weighting to produce robust Top-n matches, while exploring exhaustive and approximate search strategies for speed-accuracy trade-offs. The work introduces a 16-language multilingual benchmark and analyzes a million-scale MegaGeoCOV dataset to reveal global patterns in crisis communication. Findings show CrisisTransformers-based classifiers outperform baselines, and the TTS approach markedly improves cross-lingual matching, with practical implications for scalable, multilingual disaster relief coordination. The study lays groundwork for future benchmarks, dataset expansion, and geolocation-independent matching approaches in crisis informatics.

Abstract

During times of crisis, social media platforms play a crucial role in facilitating communication and coordinating resources. In the midst of chaos and uncertainty, communities often rely on these platforms to share urgent pleas for help, extend support, and organize relief efforts. However, the overwhelming volume of conversations during such periods can escalate to unprecedented levels, necessitating the automated identification and matching of requests and offers to streamline relief operations. Additionally, there is a notable absence of studies conducted in multi-lingual settings, despite the fact that any geographical area can have a diverse linguistic population. Therefore, we propose CReMa (Crisis Response Matcher), a systematic approach that integrates textual, temporal, and spatial features to address the challenges of effectively identifying and matching requests and offers on social media platforms during emergencies. Our approach utilizes a crisis-specific pre-trained model and a multi-lingual embedding space. We emulate human decision-making to compute temporal and spatial features and non-linearly weigh the textual features. The results from our experiments are promising, outperforming strong baselines. Additionally, we introduce a novel multi-lingual dataset simulating help-seeking and offering assistance on social media in 16 languages and conduct comprehensive cross-lingual experiments. Furthermore, we analyze a million-scale geotagged global dataset to understand patterns in seeking help and offering assistance on social media. Overall, these contributions advance the field of crisis informatics and provide benchmarks for future research in the area.

Figures (4)

• Figure 1: A high-level overview of the proposed approach for matching requests and offers shared on social media.
• Figure 2: Illustrations of language distributions (in %, left: request texts, right: offer texts) across random datasets.
• Figure 3: Global distribution of request and offer tweets during the COVID-19 pandemic. Countries considered: # geotagged tweets $>$ 50.
• Figure 4: Offers to requests ratio ($O:R$) of countries considered in Figure \ref{['global-dist']}.