1Cademy - Evaluation-First Framing of Graph-Aware RAG Comparisons (Adopted from Zeng 2025, Han 2025, Ju 2025)

Learn Before

Unbiased GraphRAG Evaluation Framework (Zeng et al., 2025)
RAG vs Graph-RAG Controlled Comparison (Han et al., 2025)
CRUX Controlled RAG Context Evaluation (Ju et al., 2025)
Scope Part 1: Strict-Parity Evaluation Contract for Graph-Aware (Auditable Strict-Parity Graph-RAG Paper)
CRUX: Controlled Retrieval-Augmented Context Evaluation for Long-Form RAG (Ju et al., 2025)
RAG And Graph-Aware Retrieval (Related Work) in Auditable Strict-Parity Evaluation of Prerequisite-Graph Retrieval for RAG under Leakage Controls

Concept

Evaluation-First Framing of Graph-Aware RAG Comparisons (Adopted from Zeng 2025, Han 2025, Ju 2025)

The paper adopts an evaluation-first framing for graph-aware RAG comparisons: recent work shows that graph-aware RAG results are sensitive to protocol choices and benefit from controlled context evaluation. Concretely, the paper compares flat dense retrieval, graph baselines, reranking, diffusion, and best-first traversal only under matched encoder, candidate-pool, cutoff, and split settings for its headline claims. This framing is taken from three prior evaluation-focused works: Zeng et al. (2025) on unbiased GraphRAG evaluation, Han et al. (2025) on controlled RAG-vs-GraphRAG comparison, and Ju et al. (2025) on controlled retrieval-augmented context evaluation (CRUX). The paper's own complementary scope — leakage-controlled prerequisite QA, question- and target-disjoint split audits, token-cap diagnostics, and a claim-to-artifact traceability layer over each headline result — is layered on top of this matched-condition framing.

0

1

Updated 2026-05-17

Contributors are: