Risk Categorization

SESAMm utilizes specialized SetFit models, trained individually on Environmental, Social, and Governance (ESG) pillars, to accurately classify risks and sub-risks. To ensure deep context, the categorization builds iteratively from individual documents up to the overarching case:

• Document-level classification: the models first analyze the combined 'main story' and 'novelty' summaries of each article to detect specific sub-risks.
• Event-level aggregation: sub-risks are assigned to an event using majority aggregation. a sub-risk must be present in at least 50% of the event's underlying documents to be validated.
• Case-level aggregation: cases aggregate all sub-risks from their constituent events. They are displayed in order of total probability.

Risk and sub-risk categorization may evolve over time if new, more severe or more reliable articles emerge.