Abstract
Official monitoring systems frequently lag behind early warning cues that emerge in advisories, news articles, and social media. These sources hold subtle indicators of change, yet their unstructured and fragmented nature makes early integration difficult. To bridge this gap, a structured pipeline was created that operates automatically or on demand, collecting raw online text from multiple open sources. Each entry is standardized into a unified schema where provenance is preserved and duplicates are removed through deterministic content hashing. The refined corpus is then examined using non-intrusive Large Language Model (LLM) queries. After clustering, the system labels, summarizes, and generates interactive Q&A responses linked to quoted snippets and verified URLs. In one example, the pipeline processed content from Google Search, online news feeds, Twitter/X, and Reddit. It clearly separated H5N1 health alerts from general advisories and revealed small yet meaningful cues such as Durango-specific confirmations and biosecurity terms.
Recommended Citation
Sheikh, Shakeel; Jain, Aarushi; and Shah, Christina Sanchita, "Human-AI collaboration in Early Warning Systems: Enhancing
decision making with Weak Signals" (2025). ACIS 2025 Proceedings. 119.
https://aisel.aisnet.org/acis2025/119