Phase D Round 1A v2 — arXiv Baseline Scan¶

掃描時間：2026-06-11 03:09

arXiv 候選數：1000 篇（日期：2023-06-01 → 2026-06-11）

Top N：30（其中 0 篇被 exclude 懲罰後仍進榜）

排序方式（v2）¶

語意相似度（cosine on nomic-embed）→ exclude 三級懲罰 → Claude Haiku R/Q/S 評分重排（final = 0.6R + 0.25S + 0.15Q）。

Tier	命中內容	係數
1 典範衝突	global GraphRAG / community detection / Leiden	×0.3
2 方法論不採用	LoRA・QLoRA・parameter-efficient / continual pretraining	×0.5
3 其他排除	chatbot UI / hiring / careers 等 off-topic	×0.7

各階段狀態：

❌ Stage 4 規則加權（venue tier、citation、recency）
✅ Stage 5 Claude Haiku 評分（R / Q / S 三維度，transitive-aware）
❌ Stage 6 Claude Sonnet 深度分析（引用建議、差異化論述）
❌ Stage 7 VPN 下載清單（IEEE/Springer 等付費期刊）

高 similarity ≠ 必然相關（embedding 可能誤判）。請當作 初步篩選，後續會由 Claude 評分修正。

Top 30 候選論文¶

1. S-Path-RAG: Semantic-Aware Shortest-Path Retrieval Augmented Generation for Multi-Hop Knowledge Graph Question Answering¶

相似度：0.7103
Haiku 評分：final 9.25 (R=9 Q=9 S=10)｜核心聚焦 Definition B：透過 k-shortest path enumeration + semantic weighting 在知識圖譜上進行 transitive multi-hop inference，不依賴向量檢索彙整，直接推導隱含關係，並有迭代圖編輯機制優化推理路徑，與本論文 GraphRAG+KG transitive reasoning 主張完全契合。
作者：Rong Fu, Yemin Wang, Tianxiang Xu et al.
發表日期：2026-03-05
arXiv：2603.23512
PDF：https://arxiv.org/pdf/2603.23512v1

We present S-Path-RAG, a semantic-aware shortest-path Retrieval-Augmented Generation framework designed to improve multi-hop question answering over large knowledge graphs. S-Path-RAG departs from one-shot, text-heavy retrieval by enumerating bounded-length, semantically weighted candidate paths using a hybrid weighted $k$-shortest, beam, and constrained random-walk strategy, learning a differentiable path scorer together with a contrastive path encoder and lightweight verifier, and injecting a ...

2. Toward Robust GraphRAG: Mitigating Retrieval Drift and Hallucination from Imperfect Knowledge Graphs¶

相似度：0.7338
Haiku 評分：final 8.85 (R=9 Q=8 S=9)｜直接聚焦於 GraphRAG 中的 transitive multi-hop reasoning（Definition B），明確識別 spurious noise 與 incomplete information 如何破壞圖譜走訪推導，並提出「sufficiency check」與「textual recovery」機制來補強不完整的隱含關係鏈，正面挑戰 LLM-KG 的結構可靠性假設，與本論文 KG-based evidence chaining 高度相關且有重要差異化視角。
作者：Yizhuo Ma, Jinchuan Xu, Tao Wen et al.
發表日期：2026-03-16
arXiv：2603.14828
PDF：https://arxiv.org/pdf/2603.14828v2

Graph Retrieval-Augmented Generation (GraphRAG) has become a common approach for multi-hop reasoning by using knowledge graphs (KGs) as structured retrieval indexes. However, most existing GraphRAG methods implicitly assume that LLM-constructed KGs provide structural support for evidence chaining. In this paper, we show that this assumption does not always hold in practice through an empirical analysis, and identify two recurring KG issue modes often overlooked by current retrievers: spurious no...

3. SciNets: Graph-Constrained Multi-Hop Reasoning for Scientific Literature Synthesis¶

相似度：0.7015
Haiku 評分：final 8.85 (R=9 Q=8 S=9)｜核心正是Definition B——通過文獻衍生的概念圖進行graph traversal，識別單篇論文中不直接共現的概念間的transitive推導路徑，明確區分graph constrained multi-hop reasoning與向量檢索，並系統比較不同圖遍歷策略（shortest-path、k-shortest、random walk）對推理深度與穩定性的影響，直接解決論文GraphRAG中的transitive relationship推導問題。
作者：Sauhard Dubey
發表日期：2025-12-28
arXiv：2601.09727
PDF：https://arxiv.org/pdf/2601.09727v1

Cross-domain scientific synthesis requires connecting mechanistic explanations across fragmented literature, a capability that remains challenging for both retrieval-based systems and unconstrained language models. While recent work has applied large language models to scientific summarization and question answering, these approaches provide limited control over reasoning depth and structural grounding. We frame mechanistic synthesis as a graph-constrained multi-hop reasoning problem over litera...

4. HyperGraphPro: Progress-Aware Reinforcement Learning for Structure-Guided Hypergraph RAG¶

相似度：0.6972
Haiku 評分：final 8.85 (R=9 Q=8 S=9)｜核心正視 Definition B transitive multi-hop reasoning，提出structure-aware hypergraph retrieval與graph traversal的密集獎勵機制，直接針對隱含關係推導的多跳圖遍歷問題，與論文GraphRAG改善transitive reasoning的claim高度對齊；實驗充分但需更明確區分transitive vs彙整型baseline。
作者：Jinyoung Park, Sanghyeok Lee, Omar Zia Khan et al.
發表日期：2026-01-25
arXiv：2601.17755
PDF：https://arxiv.org/pdf/2601.17755v2

Graph Retrieval-Augmented Generation (GraphRAG) has emerged as a promising paradigm that organizes external knowledge into structured graphs of entities and relations, enabling large language models (LLMs) to perform complex reasoning beyond text-chunk retrieval. Recent advances have integrated reinforcement learning (RL) into agentic GraphRAG approaches, enabling iterative interactions with knowledge graphs during training. However, existing RL-based methods suffer from two key limitations: (1)...

5. The Reasoning Bottleneck in Graph-RAG: Structured Prompting and Context Compression for Multi-Hop QA¶

相似度：0.6997
Haiku 評分：final 8.40 (R=8 Q=9 S=9)｜直接針對multi-hop QA的推理瓶頸，採用SPARQL分解與圖遍歷來進行transitive multi-hop推導，並在三個標準基準上驗證，方法嚴謹、baseline充分，對GraphRAG系統的結構化推理和定義B類型的跳轉推導具有重要參考價值。
作者：Yasaman Zarrinkia, Venkatesh Srinivasan, Alex Thomo
發表日期：2026-03-14
arXiv：2603.14045
PDF：https://arxiv.org/pdf/2603.14045v2

Graph-RAG systems achieve strong multi-hop question answering by indexing documents into knowledge graphs, but strong retrieval does not guarantee strong answers. Evaluating KET-RAG, a leading Graph-RAG system, on three multi-hop QA benchmarks (HotpotQA, MuSiQue, 2WikiMultiHopQA), we find that 77% to 91% of questions have the gold answer in the retrieved context, yet accuracy is only 35% to 78%, and 73% to 84% of errors are reasoning failures. We propose two augmentations: (i) SPARQL chain-of-th...

6. Beyond Chunk-Local Extraction: Cross-Chunk Graph Augmentation for GraphRAG¶

相似度：0.7443
Haiku 評分：final 8.25 (R=8 Q=8 S=9)｜核心在跨 chunk 的隱含關係補全（cross-chunk relations），明確對應 Definition B 的 transitive multi-hop reasoning，但未完全明確區分 transitive 推導 vs 資訊彙整，是本論文直接補強 GraphRAG 知識圖構建弱點的高度相關工作。
作者：Jiaming Zhang, Yibo Zhao, Jing Yu et al.
發表日期：2026-05-27
arXiv：2605.28004
PDF：https://arxiv.org/pdf/2605.28004v1

GraphRAG extends retrieval-augmented generation by organizing corpora as explicit knowledge graphs, enabling graph-based retrieval for complex question answering. However, existing frameworks extract entities and relations within individual chunks, leaving cross-chunk relations -- those whose evidence spans multiple passages -- systematically absent from the index. Exhaustive LLM-based recovery of such relations is impractical due to the combinatorial explosion of chunk combinations. We present ...

7. STEM: Structure-Tracing Evidence Mining for Knowledge Graphs-Driven Retrieval-Augmented Generation¶

相似度：0.7028
Haiku 評分：final 8.25 (R=8 Q=8 S=9)｜明確處理多跳推理（multi-hop reasoning）且採用圖遍歷策略（schema-guided graph search），直接針對隱含關係需逐跳推導的Definition B場景，但未明確區分transitive推導vs資訊彙整；Triple-GNN與全局結構指導強化了圖遍歷而非向量檢索，具高相關性與可引用性。
作者：Peng Yu, En Xu, Bin Chen et al.
發表日期：2026-04-24
arXiv：2604.22282
PDF：https://arxiv.org/pdf/2604.22282v2

Knowledge Graph-based Question Answering (KGQA) plays a pivotal role in complex reasoning tasks but remains constrained by two persistent challenges: the structural heterogeneity of Knowledge Graphs(KGs) often leads to semantic mismatch during retrieval, while existing reasoning path retrieval methods lack a global structural perspective. To address these issues, we propose Structure-Tracing Evidence Mining (STEM), a novel framework that reframes multi-hop reasoning as a schema-guided graph sear...

8. ReGraM: Region-First Knowledge Graph Reasoning for Medical Question Answering¶

相似度：0.6848
Haiku 評分：final 8.25 (R=8 Q=8 S=9)｜核心明確處理 multi-hop reasoning over 醫療 KG，通過 query-aligned subgraph 構造和逐跳推理來解決隱含關係推導問題，符合 Definition B 的 transitive reasoning 精神；實驗充分、baseline 合理、ablation 詳盡；對本論文有高價值——可引用其『區域優先』策略改進 GraphRAG 的檢索精度，並強化 multi-hop transitive 與簡單彙整的區分論述。
作者：Chaerin Lee, Sohee Park, Hyunsik Na et al.
發表日期：2026-01-14
arXiv：2601.09280
PDF：https://arxiv.org/pdf/2601.09280v1

Recent studies in medical question answering (Medical QA) have actively explored the integration of large language models (LLMs) with biomedical knowledge graphs (KGs) to improve factual accuracy. However, most existing approaches still rely on traversing the entire KG or performing large-scale retrieval, which introduces substantial noise and leads to unstable multi-hop reasoning. We argue that the core challenge lies not in expanding access to knowledge, but in identifying and reasoning over t...

9. BridgeRAG: Training-Free Bridge-Conditioned Retrieval for Multi-Hop Question Answering¶

相似度：0.7060
Haiku 評分：final 8.15 (R=8 Q=9 S=8)｜BridgeRAG明確處理transitive multi-hop reasoning（Definition B）：後續跳躍的證據需依條件於已檢索的bridge evidence重排，而非單純向量相似度，體現了圖遍歷推導的核心思想，雖無顯式KG但通過bridge-conditioned LLM judge實現隱含關係推導，方法嚴謹、實驗控制充分，可引用補強論文的推導機制與訓練自由方案。
作者：Andre Bacellar
發表日期：2026-04-03
arXiv：2604.03384
PDF：https://arxiv.org/pdf/2604.03384v2

Multi-hop retrieval is not a single-step relevance problem: later-hop evidence should be ranked by its utility conditioned on retrieved bridge evidence, not by similarity to the original query alone. We present BridgeRAG, a training-free, graph-free retrieval method for retrieval-augmented generation (RAG) over multi-hop questions that operationalizes this view with a tripartite scorer s(q,b,c) over (question, bridge, candidate). BridgeRAG separates coverage from scoring: dual-entity ANN expansi...

10. HyperRAG: Reasoning N-ary Facts over Hypergraphs for Retrieval Augmented Generation¶

相似度：0.7032
Haiku 評分：final 8.15 (R=8 Q=9 S=8)｜核心聚焦n-ary hypergraph上的多跳推理，通過結構-語義推理學習query-conditioned relational chains，明確實現沿圖譜的transitive multi-hop reasoning (Definition B)；結合adaptive high-order traversal與interpretable推理機制，方法嚴謹實驗充分，但未明確區分transitive inference vs 資訊彙整，且hypergraph相比binary KG的差異化優勢論述可深化。
作者：Wen-Sheng Lien, Yu-Kai Chan, Hao-Lung Hsiao et al.
發表日期：2026-02-16
arXiv：2602.14470
PDF：https://arxiv.org/pdf/2602.14470v1

Graph-based retrieval-augmented generation (RAG) methods, typically built on knowledge graphs (KGs) with binary relational facts, have shown promise in multi-hop open-domain QA. However, their rigid retrieval schemes and dense similarity search often introduce irrelevant context, increase computational overhead, and limit relational expressiveness. In contrast, n-ary hypergraphs encode higher-order relational facts that capture richer inter-entity dependencies and enable shallower, more efficien...

11. Retrieving Minimal and Sufficient Reasoning Subgraphs with Graph Foundation Models for Path-aware GraphRAG¶

相似度：0.7007
Haiku 評分：final 8.15 (R=8 Q=9 S=8)｜核心聚焦 multi-hop path-aware reasoning 與 subgraph retrieval，明確採用圖結構遍歷抽取 reasoning paths（Definition B transitive reasoning），且針對 cold-start 冷啟動情境提出 graph foundation model 跨域檢索方案；方法嚴謹（Information Bottleneck 優化、golden core set 抽取）、實驗充分，但未明確區分與 Definition A 彙整型的邊界，以及對山羊疾病 domain 特定性的適配程度。
作者：Haonan Yuan, Qingyun Sun, Junhua Shi et al.
發表日期：2026-03-07
arXiv：2603.07179
PDF：https://arxiv.org/pdf/2603.07179v1

Graph-based retrieval-augmented generation (GraphRAG) exploits structured knowledge to support knowledge-intensive reasoning. However, most existing methods treat graphs as intermediate artifacts, and the few subgraph-based retrieval methods depend on heuristic rules coupled with domain-specific distributions. They fail in typical cold-start scenarios where data in target domains is scarce, thus yielding reasoning contexts that are either informationally incomplete or structurally redundant. In ...

12. GRASP: Graph Agentic Search over Propositions for Multi-hop Question Answering¶

相似度：0.6867
Haiku 評分：final 8.15 (R=8 Q=9 S=8)｜GRASP 核心處理 Definition B（transitive multi-hop）的圖遍歷推導：透過 entity-proposition-passage 三層階層圖進行結構化跨跳推理，動態規畫多代理協調以解決隱含關係查詢；相比 RAG 的向量彙整，提供了明確的圖遍歷機制、token 效率優化與多資料集驗證，對 GraphRAG 框架的推導能力強化有具體貢獻價值。
作者：Stockton Jenkins, Ramya Korlakai Vinayak, Junjie Hu
發表日期：2026-05-15
arXiv：2605.16598
PDF：https://arxiv.org/pdf/2605.16598v1

Agentic retrieval improves multi-hop question answering by giving language models autonomy to iteratively gather evidence. Recent work augments these systems with knowledge graphs for structured traversal, but this combination introduces significant cost: expensive graph construction at index time and compounding token usage at inference time. We introduce Graph Agentic Search over Propositions (GRASP), an agentic system that simultaneously optimizes for high accuracy and minimal token usage in ...

13. KG-Hopper: Empowering Compact Open LLMs with Knowledge Graph Reasoning via Reinforcement Learning¶

相似度：0.6847
Haiku 評分：final 8.15 (R=8 Q=9 S=8)｜核心處理 Definition B 的 transitive multi-hop KG reasoning（整合式端到端 KG traversal），通過 RL 在單次推理中實現全局圖遍歷和動態路徑探索，直接針對「隱含關係需逐跳推導」的問題；方法嚴謹、8 個基準測試充分、相較多步驟系統展示顯著優勢，為 transitive-aware GraphRAG 框架提供有力的 RL 替代方案，可引用補強本論文端到端推理策略。
作者：Shuai Wang, Yinan Yu
發表日期：2026-03-22
arXiv：2603.21440
PDF：https://arxiv.org/pdf/2603.21440v4

Large Language Models (LLMs) demonstrate impressive natural language capabilities but often struggle with knowledge-intensive reasoning tasks. Knowledge Base Question Answering (KBQA), which leverages structured Knowledge Graphs (KGs) exemplifies this challenge due to the need for accurate multi-hop reasoning. Existing approaches typically perform sequential reasoning steps guided by predefined pipelines, restricting flexibility and causing error cascades due to isolated reasoning at each step. ...

14. GATHER: Convergence-Centric Hyper-Entity Retrieval for Zero-Shot Cell-Type Annotation¶

相似度：0.6837
Haiku 評分：final 8.15 (R=8 Q=9 S=8)｜GATHER 核心解決多基因(hyper-entity)聯合推導的多跳推理,通過圖遍歷找到收斂點(convergence nodes)來捕捉實體協同效應,屬 Definition B 的隱含transitive推導,而非單純資訊彙整,具有明確的圖遍歷優勢與嚴謹的KG-RAG baseline對比,可直接補強本論文GraphRAG在複雜多源推理場景的論述。
作者：Zhonghui Zhang, Feng Jiang, Shaowei Qin et al.
發表日期：2026-05-07
arXiv：2605.06403
PDF：https://arxiv.org/pdf/2605.06403v1

Zero-shot single-cell cell-type annotation aims to determine a cell's type from a given set of expressed genes without any training. Existing knowledge-graph-based RAG approaches retrieve evidence by expanding from source entities and relying on iterative LLM reasoning. However, in this setting each query contains tens to hundreds of genes, where no single gene is decisive and the label emerges only from their collective co-occurrence. Such hyper-entity queries fundamentally challenge local, ent...

15. HELP: HyperNode Expansion and Logical Path-Guided Evidence Localization for Accurate and Efficient GraphRAG¶

相似度：0.7592
Haiku 評分：final 8.00 (R=8 Q=8 S=8)｜核心在於改善multi-hop推理的圖譜遍歷效率，HyperNode將多跳triplet鏈接成推導路徑(transitive Definition B)，通過邏輯路徑映射避免昂貴隨機遊走；明確處理隱含關係的圖遍歷推導，雖未刻意區分Definition A/B但方法設計體現transitive推理價值。
作者：Yuqi Huang, Ning Liao, Kai Yang et al.
發表日期：2026-02-24
arXiv：2602.20926
PDF：https://arxiv.org/pdf/2602.20926v1

Large Language Models (LLMs) often struggle with inherent knowledge boundaries and hallucinations, limiting their reliability in knowledge-intensive tasks. While Retrieval-Augmented Generation (RAG) mitigates these issues, it frequently overlooks structural interdependencies essential for multi-hop reasoning. Graph-based RAG approaches attempt to bridge this gap, yet they typically face trade-offs between accuracy and efficiency due to challenges such as costly graph traversals and semantic nois...

16. Relink: Constructing Query-Driven Evidence Graph On-the-Fly for GraphRAG¶

相似度：0.7044
Haiku 評分：final 8.00 (R=8 Q=8 S=8)｜該論文核心解決 KG 不完整導致推導路徑斷裂的問題，透過動態構建 query-specific 証據圖來修復 multi-hop 推導路徑(transitive reasoning over graph)，並過濾干擾事實，直接面向 Definition B 的 graph traversal 隱含推導，但未明確強調 transitive 特性與 Definition A 的區分。
作者：Manzong Huang, Chenyang Bu, Yi He et al.
發表日期：2026-01-12
arXiv：2601.07192
PDF：https://arxiv.org/pdf/2601.07192v1

Graph-based Retrieval-Augmented Generation (GraphRAG) mitigates hallucinations in Large Language Models (LLMs) by grounding them in structured knowledge. However, current GraphRAG methods are constrained by a prevailing \textit{build-then-reason} paradigm, which relies on a static, pre-constructed Knowledge Graph (KG). This paradigm faces two critical challenges. First, the KG's inherent incompleteness often breaks reasoning paths. Second, the graph's low signal-to-noise ratio introduces distrac...

17. A Navigational Approach for Comprehensive RAG via Traversal over Proposition Graphs¶

相似度：0.6860
Haiku 評分：final 8.00 (R=8 Q=8 S=8)｜直接針對 multi-hop QA over 異質圖（proposition graphs）的結構化遍歷，明確在複雜查詢中進行 transitive inference；雖未明確標籤定義 A/B，但「缺乏結構連通性導致失敗」暗示 Definition B；實驗涵蓋簡單、複雜、抽象 QA 三類；可作為論文中 GraphRAG 與 KG-RAG 融合的具體對比方案。
作者：Maxime Delmas, Lei Xu, André Freitas
發表日期：2026-01-08
arXiv：2601.04859
PDF：https://arxiv.org/pdf/2601.04859v1

Standard RAG pipelines based on chunking excel at simple factual retrieval but fail on complex multi-hop queries due to a lack of structural connectivity. Conversely, initial strategies that interleave retrieval with reasoning often lack global corpus awareness, while Knowledge Graph (KG)-based RAG performs strongly on complex multi-hop tasks but suffers on fact-oriented single-hop queries. To bridge this gap, we propose a novel RAG framework: ToPG (Traversal over Proposition Graphs). ToPG model...

18. UltRAG: a Universal Simple Scalable Recipe for Knowledge Graph RAG¶

相似度：0.7065
Haiku 評分：final 7.85 (R=8 Q=7 S=8)｜直接針對KG上的多跳推理(multi-hop reasoning)，採用neural query executor模組沿圖遍歷執行，屬Definition B的transitive推導場景，但未明確區分transitive vs彙整型推理，且重點偏向可擴展性與成本而非推導機制本身。
作者：Dobrik Georgiev, Kheeran Naidu, Alberto Cattaneo et al.
發表日期：2026-01-28
arXiv：2603.28773
PDF：https://arxiv.org/pdf/2603.28773v1

Large language models (LLMs) frequently generate confident yet factually incorrect content when used for language generation (a phenomenon often known as hallucination). Retrieval augmented generation (RAG) tries to reduce factual errors by identifying information in a knowledge corpus and putting it in the context window of the model. While this approach is well-established for document-structured data, it is non-trivial to adapt it for Knowledge Graphs (KGs), especially for queries that requir...

19. PersonalAI 2.0: Enhancing knowledge graph traversal/retrieval with planning mechanism for Personalized LLM Agents¶

相似度：0.7021
Haiku 評分：final 7.85 (R=8 Q=7 S=8)｜該論文明確針對多跳推理開發圖遍歷演算法(BeamSearch、WaterCircles)和規劃機制,在HotpotQA、2WikiMultihopQA等多跳基準上展示改進,但未明確區分transitive隱含關係推導vs多chunk彙整,且為通用PersonalAI框架而非專注於transitive Definition B的核心機制,可作為GraphRAG改進方向參考但差異化論述有限。
作者：Mikhail Menschikov, Matvey Iskornev, Alexander Kharitonov et al.
發表日期：2026-05-13
arXiv：2605.13481
PDF：https://arxiv.org/pdf/2605.13481v1

We introduce PersonalAI 2.0 (PAI-2), a novel framework, designed to enhance large language model (LLM) based systems through integration of external knowledge graphs (KG). The proposed approach addresses key limitations of existing Graph Retrieval-Augmented Generation (GraphRAG) methods by incorporating a dynamic, multistage query processing pipeline. The central point of PAI-2 design is its ability to perform adaptive, iterative information search, guided by extracted entities, matched graph ve...

20. STAR: Semantic-Tuned and Tail-Adaptive Retriever for Graph-Augmented Generation¶

相似度：0.7454
Haiku 評分：final 7.75 (R=8 Q=8 S=7)｜直接處理GraphRAG的多跳KG推理，透過路徑加權與token級互動學習改進檢索，符合transitive multi-hop需求；但未明確區分Definition B（圖遍歷推導隱含關係）vs Definition A（彙整型），且未量化multi-hop的實際跳數或transitive推理佔比。
作者：Shuai Li, Chen Huang, Duanyu Feng et al.
發表日期：2026-04-11
arXiv：2605.18765
PDF：https://arxiv.org/pdf/2605.18765v1

To augment Large Language Models (LLMs) for multi-hop question answering, a mainstream solution within Graph Retrieval Augmented Generation (GraphRAG) leverages lightweight retrievers to efficiently extract information from a given Knowledge Graph (KG). However, existing methods often overlook the inherent challenge of sparse semantic information in graphs. Specifically, our experiments reveal that these methods produce biased retrieval Semantic Shortcut Bias and Long-Tail Path Bias, leading to ...

21. Do We Still Need GraphRAG? Benchmarking RAG and GraphRAG for Agentic Search Systems¶

相似度：0.7411
Haiku 評分：final 7.75 (R=8 Q=8 S=7)｜明確處理 multi-hop reasoning 並評估 GraphRAG 優勢，但未清楚區分 transitive 推導型 vs 多 chunk 彙整型 multi-hop，主要貢獻在 agentic search 補償能力而非 Definition B 的 graph traversal 顯著證據。
作者：Dongzhe Fan, Zheyi Xue, Siyuan Liu et al.
發表日期：2026-04-01
arXiv：2604.09666
PDF：https://arxiv.org/pdf/2604.09666v1

Retrieval-augmented generation (RAG) and its graph-based extensions (GraphRAG) are effective paradigms for improving large language model (LLM) reasoning by grounding generation in external knowledge. However, most existing RAG and GraphRAG systems operate under static or one-shot retrieval, where a fixed set of documents is provided to the LLM in a single pass. In contrast, recent agentic search systems enable dynamic, multi-round retrieval and sequential decision-making during inference, and h...

22. BubbleRAG: Evidence-Driven Retrieval-Augmented Generation for Black-Box Knowledge Graphs¶

相似度：0.7198
Haiku 評分：final 7.75 (R=8 Q=8 S=7)｜處理black-box KG上的multi-hop QA，採graph traversal方式(semantic anchor、bubble expansion)發現candidate evidence subgraphs，核心聚焦於結構路徑不確定性(structural path uncertainty)的圖譜推導，符合Definition B的transitive reasoning需求；惟未明確區分與多chunk彙整型的差異，且主要貢獻在recall/precision框架設計而非transitive性質本身。
作者：Duyi Pan, Tianao Lou, Xin Li et al.
發表日期：2026-03-19
arXiv：2603.20309
PDF：https://arxiv.org/pdf/2603.20309v1

Large Language Models (LLMs) exhibit hallucinations in knowledge-intensive tasks. Graph-based retrieval augmented generation (RAG) has emerged as a promising solution, yet existing approaches suffer from fundamental recall and precision limitations when operating over black-box knowledge graphs -- graphs whose schema and structure are unknown in advance. We identify three core challenges that cause recall loss (semantic instantiation uncertainty and structural path uncertainty) and precision los...

23. Beyond Explicit Edges: Robust Reasoning over Noisy and Sparse Knowledge Graphs¶

相似度：0.7062
Haiku 評分：final 7.75 (R=8 Q=8 S=7)｜核心針對 KG 稀疏/雜訊下的 multi-hop reasoning，藉由 LLM 導航與 embedding 相似度擴展來推導隱含關係，符合 Definition B (transitive inference via graph traversal)，但未明確區分 transitive vs 彙整型，且重點在魯棒性而非純粹 transitive 推導的形式化驗證。
作者：Hang Gao, Dimitris N. Metaxas
發表日期：2026-03-14
arXiv：2603.14006
PDF：https://arxiv.org/pdf/2603.14006v1

GraphRAG is increasingly adopted for converting unstructured corpora into graph structures to enable multi-hop reasoning. However, standard graph algorithms rely heavily on static connectivity and explicit edges, often failing in real-world scenarios where knowledge graphs (KGs) are noisy, sparse, or incomplete. To address this limitation, we introduce INSES (Intelligent Navigation and Similarity Enhanced Search), a dynamic framework designed to reason beyond explicit edges. INSES couples LLM-gu...

24. SCPRM: A Schema-aware Cumulative Process Reward Model for Knowledge Graph Question Answering¶

相似度：0.6974
Haiku 評分：final 7.75 (R=8 Q=8 S=7)｜直接處理 KG 上的 multi-hop reasoning，明確區分多路徑且強調「錯誤步驟導致推導路徑失效」(Definition B transitive)；MCTS+schema distance 設計針對圖遍歷的路徑評分，但未明確區分 transitive inference 與彙整型，且改進幅度溫和。
作者：Jiujiu Chen, Yazheng Liu, Sihong Xie et al.
發表日期：2026-05-04
arXiv：2605.02819
PDF：https://arxiv.org/pdf/2605.02819v1

Large language models excel at complex reasoning, yet evaluating their intermediate steps remains challenging. Although process reward models provide step-wise supervision, they often suffer from a risk compensation effect, where incorrect steps are offset by later correct ones, assigning high rewards to flawed reasoning paths. This issue is further exacerbated in knowledge graph (KG) reasoning, as there may exist multiple paths between the start and end entities in the KGs, and a risky step can...

25. IndexRAG: Bridging Facts for Cross-Document Reasoning at Index Time¶

相似度：0.7141
Haiku 評分：final 7.40 (R=7 Q=8 S=8)｜處理 multi-hop QA 跨文件推理，核心利用 bridge entities 在索引時預生成銜接事實，減少在線圖遍歷成本；但未明確區分是否涉及 transitive 隱含關係推導(Definition B)或主要是多文件事實彙整(Definition A)，屬圖強化 RAG 範疇且有充分實驗基礎，對本論文的 GraphRAG 框架有比較價值。
作者：Zhenghua Bao, Yi Shi
發表日期：2026-03-17
arXiv：2603.16415
PDF：https://arxiv.org/pdf/2603.16415v1

Multi-hop question answering (QA) requires reasoning across multiple documents, yet existing retrieval-augmented generation (RAG) approaches address this either through graph-based methods requiring additional online processing or iterative multi-step reasoning. We present IndexRAG, a novel approach that shifts cross-document reasoning from online inference to offline indexing. IndexRAG identifies bridge entities shared across documents and generates bridging facts as independently retrievable u...

26. SentGraph: Hierarchical Sentence Graph for Multi-hop Retrieval-Augmented Question Answering¶

相似度：0.6933
Haiku 評分：final 7.40 (R=7 Q=8 S=8)｜處理文件間多跳QA且建圖構造logic relations，但重點在句子層級coherence與entity bridges(Definition A彙整型)，非transitive推導；圖策略合理、實驗充分，可引用補強檢索品質但不涉及隱含關係推導。
作者：Junli Liang, Pengfei Zhou, Wangqiu Zhou et al.
發表日期：2026-01-06
arXiv：2601.03014
PDF：https://arxiv.org/pdf/2601.03014v3

Traditional Retrieval-Augmented Generation (RAG) effectively supports single-hop question answering with large language models but faces significant limitations in multi-hop question answering tasks, which require combining evidence from multiple documents. Existing chunk-based retrieval often provides irrelevant and logically incoherent context, leading to incomplete evidence chains and incorrect reasoning during answer generation. To address these challenges, we propose SentGraph, a sentence-l...

27. LogosKG: Hardware-Optimized Scalable and Interpretable Knowledge Graph Retrieval¶

相似度：0.7265
Haiku 評分：final 7.15 (R=7 Q=8 S=7)｜LogosKG 處理 k-hop KG retrieval 並優化 multi-hop traversal 效率，展示 KG topology 對 KG-LLM 推理的影響，屬 Definition B 之圖遍歷範疇；惟論文未明確區分 transitive 隱含關係推導 vs 多 chunk 資訊彙整，且主要貢獻聚焦硬體最佳化而非 transitive reasoning 本身。
作者：He Cheng, Yifu Wu, Saksham Khatwani et al.
發表日期：2026-04-20
arXiv：2604.18913
PDF：https://arxiv.org/pdf/2604.18913v1

Knowledge graphs (KGs) are increasingly integrated with large language models (LLMs) to provide structured, verifiable reasoning. A core operation in this integration is multi-hop retrieval, yet existing systems struggle to balance efficiency, scalability, and interpretability. We introduce LogosKG, a novel, hardware-aligned framework that enables scalable and interpretable k-hop retrieval on large KGs by building on symbolic KG formulations and executing traversal as hardware-efficient operatio...

28. N2N-GQA: Noise-to-Narrative for Graph-Based Table-Text Question Answering Using LLMs¶

相似度：0.7217
Haiku 評分：final 7.15 (R=7 Q=8 S=7)｜處理table-text混合多跳QA且明確使用graph結構組織證據關係，但重點在於從噪音檢索結果中「識別橋接文檔」(Definition A彙整型)，而非transitive推導(Definition B)；雖有圖構造但缺乏明確的隱含關係KG遍歷機制。
作者：Mohamed Sharafath, Aravindh Annamalai, Ganesh Murugan et al.
發表日期：2026-01-10
arXiv：2601.06603
PDF：https://arxiv.org/pdf/2601.06603v1

Multi-hop question answering over hybrid table-text data requires retrieving and reasoning across multiple evidence pieces from large corpora, but standard Retrieval-Augmented Generation (RAG) pipelines process documents as flat ranked lists, causing retrieval noise to obscure reasoning chains. We introduce N2N-GQA. To our knowledge, it is the first zeroshot framework for open-domain hybrid table-text QA that constructs dynamic evidence graphs from noisy retrieval outputs. Our key insight is tha...

29. ConRAG: Consensus-Driven Multi-View Retrieval for Multi-Hop Question Answering¶

相似度：0.7162
Haiku 評分：final 7.15 (R=7 Q=8 S=7)｜處理多跳QA並運用知識圖譜的多視角檢索，但未明確區分Definition B的遞移性推導（沿圖譜多跳必需推導）與Definition A的資訊彙整型（向量檢索可解），multi-hop reasoning的核心機制未聚焦於transitive graph traversal。
作者：Yikai Zhu, Kunfeng Chen, Qihuang Zhong et al.
發表日期：2026-05-27
arXiv：2605.28093
PDF：https://arxiv.org/pdf/2605.28093v2

Retrieval-augmented generation (RAG) has emerged as a promising paradigm for enhancing large language models (LLMs) on multi-hop question answering (QA), which requires reasoning over evidence from multiple documents. Current multi-hop RAG methods generally focus on either query-side task decomposition or corpus-side knowledge graph construction. Despite their progress, these methods still struggle to achieve satisfactory performance on complex multi-hop QA tasks. To this end, we propose ConRAG,...

30. From Flat to Structural: Enhancing Automated Short Answer Grading with GraphRAG¶

相似度：0.7094
Haiku 評分：final 7.15 (R=7 Q=8 S=7)｜該論文明確針對多跳推理設計GraphRAG框架，使用圖遍歷檢索「邏輯推理鏈」(reasoning chains)，符合Definition B的transitive推導特徵；但未明確區分transitive隱含關係與多chunk彙整，NGSS評分任務焦點在知識結構而非純粹圖遍歷必要性，可引用框架與baseline對比但差異化有限。
作者：Yucheng Chu, Haoyu Han, Shen Dong et al.
發表日期：2026-02-28
arXiv：2603.19276
PDF：https://arxiv.org/pdf/2603.19276v1

Automated short answer grading (ASAG) is critical for scaling educational assessment, yet large language models (LLMs) often struggle with hallucinations and strict rubric adherence due to their reliance on generalized pre-training. While Rretrieval-Augmented Generation (RAG) mitigates these issues, standard "flat" vector retrieval mechanisms treat knowledge as isolated fragments, failing to capture the structural relationships and multi-hop reasoning essential for complex educational content. T...