Data Engineering & NLP Analysis

To overcome the challenge of unstructured text, the researchers employed an ensemble of advanced generative Large Language Models (LLMs), including Mistral-7B, Zephyr-7B, and Meta-Llama-3-70B. Instead of relying on a single model, this approach combines the outputs of multiple LLMs. A majority voting system then determines the most accurate extraction for key entities like disease name, country, date, case totals, and mortality. This ensemble method was proven to be more robust and accurate than any single open-source model and competitive with leading commercial models.

The extracted information is structured into an Epidemiological Knowledge Graph (eKG). A knowledge graph is a highly organized database that represents information as entities (like 'Ebola' or 'China') and the relationships between them (like 'occurs in'). This eKG is built on FAIR principles (Findable, Accessible, Interoperable, Reusable) and Linked Open Data (LOD) standards like RDF. This makes the data not just human-readable but also machine-interpretable, enabling complex queries, automated analysis, and seamless integration with other enterprise data systems.

For an enterprise, the eKG provides a real-time, structured feed of global health risks. This intelligence can be integrated into various systems to: 1. Fortify Supply Chains: Identify potential disruptions in manufacturing or logistics hubs. 2. Enhance Employee Safety: Issue timely travel advisories and health alerts. 3. Inform Financial Models: Quantify the potential impact of epidemics on specific markets or sectors. 4. Improve Strategic Planning: Make data-driven decisions on market entry, resource allocation, and long-term risk mitigation.