nli-score

What problem does it solve?

This Skill automates detection of contradictions between claims in a literature-derived knowledge graph, reducing manual evidence triage and producing evidence-weighted confidence metrics for each claim.

Core Features & Use Cases

• False-positive pre-filtering: skips parallel assertions and applies deterministic predicate-opposition checks to avoid spurious contradictions.
• NLI classification: runs a DeBERTa cross-encoder over claim pairs and applies Beta-Binomial updates to per-edge posteriors.
• Reporting & inspection: generates interactive HTML and JSON reports for exploration and evidence-level diagnostics.
• Use Case: score a KG built from extracted paper claims to surface contradictory findings, quantify controversy, and drive gap-aware supplementary searches.

Quick Start

Run the nli-score pipeline on your knowledge graph file to score claim contradictions and generate an interactive HTML and JSON report.