RESEARCH · RESEARCH · #1411
EviGraph: proof-carrying selective recommendation over temporal public-service knowledge graphs (arXiv:2610.00212v1)
This paper introduces EviGraph, a system that separates critical decision requirements from noncritical unknowns by having a language agent link required criteria to evidence in a temporal knowledge graph and a deterministic checker verify whether a recommendation is supported. Evaluation on a bilingual Hong Kong public-service benchmark with executable policy references shows the approach reduces unnecessary abstention, while additional verification steps can sometimes revoke supported recommendations without improving decision outcomes.
KEY POINTS
- This paper introduces EviGraph, a system that separates critical decision requirements from noncritical unknowns by having a language agent link required criteria to evidence in a temporal knowledge graph and a deterministic checker verify whether a recommendation is supported.
- Evaluation on a bilingual Hong Kong public-service benchmark with executable policy references shows the approach reduces unnecessary abstention, while additional verification steps can sometimes revoke supported recommendations without improving decision outcomes.
- Highlights that defining what must be established for a decision—rather than simply adding more verification—improves evidence-based public-service recommendations and affects abstention behavior.
WHY IT MATTERS
Highlights that defining what must be established for a decision—rather than simply adding more verification—improves evidence-based public-service recommendations and affects abstention behavior.