EviGraph: Proof-Carrying Selective Recommendation over Temporal Knowledge Graphs

Forum topic · 小凯 · 2026-10-05

Summary

EviGraph (arXiv:2610.00212) is a framework for selective recommendation over public services that distinguishes critical decision requirements from information that can remain unresolved. A language agent links each requirement to evidence stored in a temporal knowledge graph, and a deterministic checker verifies whether a recommendation is actually supported by that evidence. The system was evaluated on a bilingual Hong Kong public-service benchmark featuring executable policy references. Results show that explicitly separating decisive requirements from open questions reduces unnecessary abstention, meaning the system delivers useful recommendations instead of withholding them when only non-critical details are missing. Notably, adding extra verification steps can withdraw previously supported recommendations without improving decision quality. The authors conclude that reliable evidence-based navigation depends on precisely specifying what a decision must establish, rather than simply stacking more verification layers. The work spans machine learning, knowledge graphs, and trustworthy recommendation systems.

Paper Overview

Research area: ML
Authors: Yixi Zhou, Sikun Wang, Lei Fan, Fan Zhang
Published: 2026-10-05
arXiv: 2610.00212

Abstract

Public-service recommendations require evidence that matches the requested service, scope, and date. Yet treating every missing detail as decisive can withhold useful recommendations. We introduce EviGraph, which distinguishes critical decision requirements from information that can remain unresolved. A language agent links these requirements to evidence in a temporal knowledge graph, while a deterministic checker establishes whether a recommendation is supported. Evaluation on a bilingual Hong Kong public-service benchmark with executable policy references shows that this distinction reduces unnecessary abstention. Additional verification, however, can withdraw supported recommendations without improving decision quality. These findings suggest that reliable evidence-based navigation depends on explicitly defining what a decision must establish, rather than simply adding more verification.

Key Takeaways

Selective recommendation: EviGraph avoids withholding useful recommendations just because minor, non-decisive details are missing.
Architecture: A language agent grounds decision requirements in evidence from a temporal knowledge graph; a deterministic checker certifies whether recommendations are supported.
Benchmark: Evaluated on a bilingual Hong Kong public-service benchmark with executable policy references.
Result: The requirement/abstention distinction reduces unnecessary abstention.
Counterintuitive finding: Extra verification can retract already-supported recommendations without improving decision quality — more checking is not always better.
Implication: Trustworthy evidence-based systems should explicitly specify which facts a decision must establish.

---

*Auto-collected on 2026-10-05.*