AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Graph of Verification: Structured Verification of LLM Reasoning with Directed Acyclic Graphs

Jiwei Fang, Bin Zhang, Changwei Wang, Jin Wan, Zhiwei Xu

AAAI 2026 · Large language models › Retrieval and knowledge · approach: Knowledge graph reasoning with LLMs

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
decompose verification into atomic nodes with minimal context (GoV) instead of holistic verificationQwen2.5-72B-Instruct verifier · Number Triangle Summation (N=8)F1: 49.5 -> 98.1Improved
decompose verification into atomic nodes (GoV) instead of holisticQwen2.5-32B-Instruct verifier · Number Triangle Summation (N=2)F1: 100.0 -> 99.8No clear effect
decompose verification into atomic nodes (GoV) instead of holisticQwen2.5-72B-Instruct verifier · Number Triangle Summation (N=2)F1: 99.6 -> 98.7Hurt
paragraph-level node blocks with inclusive context (GoV) instead of holisticQwen2.5-32B-Instruct verifier · ProcessBench OlympiadBenchF1: 35.9 -> 65.2Improved
paragraph-level node blocks (GoV) instead of holisticQwen2.5-7B-Instruct verifier · ProcessBench GSM8KF1: 47.3 -> 58.0Improved
paragraph-level node blocks (GoV) instead of holisticLlama3.3-70B-Instruct verifier · ProcessBench GSM8KF1: 78.6 -> 78.6No clear effect
paragraph-level node blocks (GoV) instead of holisticLlama3.3-70B-Instruct verifier · ProcessBench MATHF1: 54.4 -> 71.5Improved
use block granularity with inclusive context (GoV) instead of PARC premise extractionQwen2.5-32B-Instruct verifier · ProcessBench GSM8KF1: 70.0 -> 78.1Improved
use block granularity with inclusive context (GoV) instead of PARCQwen2.5-72B-Instruct verifier · ProcessBench MATHF1: 66.5 -> 75.2Improved
training-free GoV with Qwen2.5-7B versus trained PRMsProcessBench · ProcessBench OlympiadBenchF1: 22.9 -> 42.9Improved
training-free GoV with Qwen2.5-7B versus Skywork-PRM-7BProcessBench · ProcessBench GSM8KF1: 70.8 -> 58.0Hurt
iterative per-node verification (GoV) instead of single-pass CoTQwen2.5-32B-Instruct verifier · Number Triangle Summation (500 instances)total cost ($): 0.140 -> 1.329Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy