Pith. sign in

REVIEW 4 major objections 5 minor 89 references

Global semantic-unit merging beats four RAG baselines in five domains

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

GOSU globally merges semantic units from text chunks into a unit-centric knowledge graph and uses three-tier keyword retrieval to improve RAG generation quality, according to LLM-judge win rates.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection GOSU has a genuinely sensible pipeline idea—globally merging chunk-local semantic units before graph construction—but the reported win rates are not interpretable because the LLM-judge protocol discards inconsistent comparisons and renormalizes without reporting counts. the 4 major comments →

arxiv 2509.00449 v1 pith:WZEFTHBY submitted 2025-08-30 cs.CL

GOSU: Retrieval-Augmented Generation with Global-Level Optimized Semantic Unit-Centric Framework

Xuecheng Zou , Ke Liu , Bingbing Wang , Huafei Deng , Li Zhang , Yu Tang This is my paper
classification cs.CL
keywords retrieval-augmented generationsemantic unitsknowledge graph constructionhypergraph relationsglobal disambiguationhierarchical retrievaln-ary relationsLLM-as-judge evaluation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes GOSU, a retrieval-augmented generation framework whose central move is to extract event-like 'semantic units' from each local text chunk, then merge and disambiguate them across the whole corpus before any graph is built. This global merging step is meant to fix a weakness in graph-based RAG: when semantic units are defined only inside isolated chunks, the same event can be split, repeated, or entangled with unrelated details, and fine-grained binary relations get lost among coarse n-ary structures. GOSU then builds a knowledge graph centered on the merged semantic units, preserving both entity-to-entity (binary) edges and multi-entity (n-ary) relations, and retrieves through a three-tier keyword scheme plus semantic-unit completion. The paper reports that GOSU wins pairwise LLM-judged comparisons against NaiveRAG, LightRAG, HiRAG, and HyperGraphRAG across Agriculture, CS, hypertension guidelines, Legal, and Mixed datasets on Comprehensiveness, Diversity, Empowerment, and Overall, with the largest average gaps against HyperGraphRAG. The claim is that globally consistent semantic units give more faithful, comprehensive, and coherent generation than either pure binary graphs or pure hypergraphs.

Core claim

GOSU's central claim is that graph-based RAG quality is set by how semantic units are defined before the graph exists. Prior methods extract events or hyperedges inside isolated chunks; GOSU collects candidate SUs from every chunk, filters them with a cosine coarse pass plus an LLM fine pass, clusters and deduplicates them corpus-wide, and only then extracts entities and binary relations guided by these merged units. The resulting graph is SU-centric: each SU node carries coarse n-ary relations while staying linked to fine-grained binary edges. Retrieval uses hierarchical keywords (entity, theme, and a new mid-level SU tier) plus SU completion, so answers draw on both pairwise facts and mult

What carries the argument

The semantic unit (SU)—a compact phrase capturing an event, fact, or concept that spans several entities—is the load-bearing object. GOSU extracts candidate SUs per chunk, filters them with cosine similarity plus an LLM judge, clusters and deduplicates them globally, then uses the merged SUs as graph nodes that anchor both binary entity–relation edges and n-ary multi-entity links. At query time a mid-level 'semantic-unit' keyword tier and an SU-completion retrieval step let the system recover coarse multi-entity events that plain entity/theme keywords would miss.

Load-bearing premise

The reported win rates all come from pairwise judgments by GPT-4o-mini, the same model that extracts the semantic units and writes the answers, and pairs where the judge flips preference with left-right order are discarded, so the remaining comparisons may be the easy ones.

What would settle it

Take the same five corpora and queries, run GOSU against HyperGraphRAG, and have human raters (or a different LLM family) judge the same four dimensions; if human preference margins fall far below the reported 70–96 percent win rates, the quality advantage is an artifact of the GPT-4o-mini judge. Alternatively, annotate multi-entity events in the corpora and measure whether globally merged SUs retrieve them with higher recall than chunk-local extraction.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If GOSU's results hold, graph RAG systems should merge event-level units across the corpus before extracting entities and relations, not after.
  • Retrieval pipelines can combine fine-grained entity/relation keywords with coarse semantic-unit completion to cover both pairwise facts and multi-entity events in one answer.
  • The semantic-unit layer contributes more to quality than either the entity or relation layer alone; removing it causes the largest drop in win rate.
  • Global merging adds offline preprocessing cost (pairwise similarity comparisons and higher embedding token use), but keeps online query cost comparable to lighter baselines.
  • GOSU's stability across five domains suggests the framework transfers across agriculture, CS, law, medicine, and mixed corpora.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the same model family (GPT-4o-mini) extracts semantic units, generates answers, and judges them, part of the reported margin may reflect self-consistency bias; a human or cross-model judge replication would clarify.
  • The pairwise SU-similarity step is a scalability bottleneck; approximate nearest-neighbor or clustering-based merging could be tested as a drop-in replacement on larger corpora.
  • The authors list multimodal input and deep multi-hop reasoning as open limitations; an extension could represent tables and images as SU-typed nodes or add reasoning-path edges between SUs.
  • A direct test: on a corpus with annotated multi-entity events, compare whether GOSU's globally merged SUs recover those events with higher recall than chunk-local extraction.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes GOSU, a graph-based retrieval-augmented generation (RAG) framework organized around "semantic units" (SUs). SUs are first extracted from local chunks, then globally merged, deduplicated, and disambiguated; the resulting SU set guides entity and relation extraction to build an SU-centric knowledge graph that preserves both fine-grained binary relations and coarse-grained n-ary relations. At retrieval and generation time, GOSU uses hierarchical keyword extraction (entity-level, semantic-unit-level, and theme-level) together with semantic-unit completion, and feeds the fused context to an LLM generator. The experimental section compares GOSU with NaiveRAG, LightRAG, HiRAG, and HyperGraphRAG on five datasets using LLM-based pairwise evaluation on Comprehensiveness, Diversity, Empowerment, and Overall dimensions, reporting very large win-rate margins (Table 1), plus an ablation study (Table 2) and a cost/efficiency analysis (Fig. 3). The central claim is that this global SU-centric design yields more faithful, comprehensive, and coherent RAG outputs than existing graph-based RAG systems.

Significance. If the empirical claim were established, GOSU would be a useful contribution to graph-based RAG: the idea of globally merging semantic units before entity/relation extraction is a plausible way to reduce coreference and fragmentation, and the two-tier retrieval design (fine-grained keywords plus coarse-grained SU completion) is interesting and moderately novel. The authors provide a public code repository, which is a strength. However, the evaluation as reported is not interpretable: the LLM-judge protocol in §4.1 discards inconsistent pairwise judgments without reporting how many are discarded, and all headline numbers in Tables 1 and 2 are conditional on that selected subset. Because the reported margins are extremely large and the protocol is biased toward inflated win rates, the evidence does not currently support the paper's central claim. The omission of the closest baselines (NodeRAG, GraphRAG) and an internal inconsistency between the described dimensions (Insightfulness) and the tables (Empowerment) further weaken the evaluation. The conceptual framework may have merit, but the empirical case needs substantial rework.

major comments (4)
  1. [§4.1, Tables 1–2] The evaluation protocol accepts a pairwise comparison only when the two alternating judgments agree; inconsistent outcomes are excluded as 'inconclusive'. Table 1 reports win rates that sum to 100% for every row, so the denominator is the accepted-consistent subset, not all trials. The paper never reports the number of queries, the number of excluded pairs, or judge agreement rates, making every number in Tables 1 and 2 unverifiable. Moreover, conditioning on consistency inflates win rates: if the judge's true preference accuracy is p>0.5, the reported win rate after requiring two consistent judgments is p^2/(p^2+(1-p)^2), which is 69% when p=0.6 and 84% when p=0.7. Thus the large margins (e.g., +88.2% vs HyperGraphRAG) may largely reflect the selection rule rather than actual quality differences. The authors should report all pairwise counts, the number/rate of excluded inconsistent jud
  2. [§4.1, Evaluation Metrics] There is an internal inconsistency in the description of the evaluation dimensions. The paragraph describing the protocol states that 'the final overall preference was determined by aggregating the rankings across the three primary dimensions (Comprehensiveness, Diversity, and Insightfulness), with ties resolved using the Overall Quality score.' However, Table 1 and Table 2 report a dimension called 'Empowerment,' not 'Insightfulness.' The two names are not interchangeable, and this discrepancy makes the evaluation protocol non-reproducible. The authors must clarify which dimensions were actually scored and whether 'Overall' is a separate judged dimension or an aggregation of the other three. This is not a typo-level issue: it affects the meaning of every row in the results tables.
  3. [§2.3, §4.1 (Baselines)] The paper's positioning relies heavily on two prior systems: NodeRAG (Xu et al., 2025b), which is described as capturing n-ary units only within isolated chunks, and GraphRAG (Edge et al., 2024), which is the standard graph-RAG baseline. Yet neither is included in the experimental comparison. Since the central claim is that GOSU improves on 'existing RAG pipelines' and specifically on chunk-local heterogeneous graphs, omitting NodeRAG—and GraphRAG, whose evaluation methodology the authors adopt—leaves the novelty claim unvalidated. The comparison set (NaiveRAG, LightRAG, HiRAG, HyperGraphRAG) is reasonable as a partial set, but the paper overstates the conclusion by calling the results a demonstration against 'the baseline RAG methods.' The authors should either add these baselines or temper the claim to the evaluated systems.
  4. [§4.1, Implementation Details; §3.1–3.3] The same model family (GPT-4o-mini) appears to be used as the semantic-unit extractor, the graph-construction LLM, the generator, and the LLM judge. If the judge is the same model that produced GOSU's outputs (and the baseline outputs are also generated by GPT-4o-mini with different prompts), the judge may systematically prefer outputs that match its own extraction and generation style. The exclusion of inconsistent comparisons compounds this risk by removing the very cases where the judge is uncertain. The authors should report judge-human agreement on a sample, use a different model as judge, or otherwise demonstrate that the judge's preferences are not an artifact of shared model family. This is a correctness-risk concern, not an allegation of misconduct.
minor comments (5)
  1. [Abstract, §1] The sentence 'The former uncovers the fine-grained binary relationships overlooked by the latter, while the latter compensates for the coarse-grained n-ary relationships missing from the former' is confusing because the antecedents of 'the former' and 'the latter' are not clear at that point. Please rephrase with explicit referents.
  2. [§3.1, Eqs. (5), (10), (22)] The symbol τ is used both as a cosine-similarity threshold in Eq. (5) and as a retrieval-budget threshold in Eqs. (10) and (22). These are different parameters and should be denoted differently.
  3. [Fig. 3 caption and §4.2] The caption writes 'TPB/TPQ: tokens per chunk/query' but the text defines TPC/CPC/TPQ/CPQ. The abbreviation TPB is not defined, and 'comparision' is a typo. Please harmonize the notation and define all abbreviations in the caption.
  4. [§4.1, Datasets] The fifth dataset is described only as 'the most recent international hypertension guidelines (McCarthy et al., 2025)'. This is a short commentary/article, not a corpus. Please specify how many documents/chunks/queries it comprises, how the corpus was constructed, and how queries were generated.
  5. [§4.2, Analysis of Efficiency and Cost] The cost/efficiency discussion is sketchy: no units are given for 'TPC reached 29,560 tokens' (is this per chunk, per corpus, or per document?), and no total end-to-end cost is reported. Clarifying the denominators would make Fig. 3 interpretable.

Circularity Check

0 steps flagged

No construction-level circularity; the paper is an empirical pipeline comparison, and its evaluation-protocol choices are concerns about validity, not circular derivation.

full rationale

GOSU is not a derivational paper: there is no chain of equations or fitted parameters that is then reported as a prediction. The framework's components—global semantic-unit merging (Eqs. 3–11), SU-centric graph construction (Eqs. 12–16), and dual-phase retrieval/generation (Eqs. 17–26)—are described procedurally and evaluated empirically against external baselines. No step defines X in terms of Y and then claims to predict Y. The strongest potentially self-referential aspect is the evaluation protocol in §4.1, which uses GPT-4o-mini as the LLM extractor/generator and as the pairwise judge and excludes inconsistent comparisons; however, this is a measurement-bias/validity concern, not a case where an output is equivalent to an input by construction. The reported win rates are direct LLM-judge measurements, not quantities derived from the method's own assumptions. The paper contains no load-bearing self-citations: the only 'Zou et al. 2025' entry in the references is Jiaru Zou's GTR paper, not the present first author, and no uniqueness theorem or ansatz is imported from the authors' own prior work. The limitations section acknowledges missing multimodal inputs, limited deep reasoning, and efficiency costs, none of which indicate circularity. Accordingly, the central claim does not reduce to its inputs, and the appropriate finding is no significant circularity (score 0).

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

No new physical or conceptual entities are postulated; semantic units are a representation adopted from prior node-based RAG work (e.g., NodeRAG). The framework's novelty lies in the pipeline, not in a new ontology.

free parameters (4)
  • Similarity threshold tau for semantic-unit filtering = not specified
    Used in Eqs. (5), (6), (10), and (22) to decide which SU pairs to merge and retrieval budgets; no value or search procedure reported.
  • Chunking parameters (size/overlap) = not specified
    DocSplit in Eqs. (1)-(2) uses length-controlled sliding windows; chunk size and overlap are not reported, though they affect local SU extraction.
  • Retrieval top-k budgets = not specified
    Trim constraints in Eqs. (10), (21), (24), (25) require preset budgets; values are not given.
  • LLM judge consistency criterion = not specified
    Section 4.1 defines 'consistent preference' but no threshold for how many annotations are needed; this criterion directly shapes reported win rates.
axioms (5)
  • domain assumption LLM-based extraction returns complete, coherent, information-bearing semantic units from each chunk.
    Eq. (3) assumes SemExtLLM(ci) produces candidate units with completeness, coherence, and information-bearing capacity; failure would invalidate the global merging.
  • domain assumption Cosine similarity plus LLM judgment reliably identifies semantically duplicate or coreferent semantic units.
    Eqs. (5)-(6) assume coarse cosine filtering and LLMJudge correctly separate duplicates from distinct events.
  • domain assumption LLM-as-judge pairwise preferences are a valid measure of generation quality.
    Section 4.1 relies on GPT-4o-mini as the evaluator for all dimensions; its reliability is not validated against human judgments.
  • domain assumption Four evaluation dimensions (Comprehensiveness, Diversity, Empowerment, Overall) capture relevant RAG quality.
    Borrowed from KnowTuning/LightRAG protocol; no justification for their sufficiency.
  • domain assumption Hyperparameters held identical across baselines.
    Section 4.1 states chunk size and retrieval hyperparameters are held identical but does not report their values or verify equivalence for graph-based pipelines.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of GOSU: Retrieval-Augmented Generation with Global-Level Optimized Semantic Unit-Centric Framework." pith.science (2026). https://pith.science/paper/WZEFTHBY

@misc{pith2026250900449,
  author       = {Pith},
  title        = {Pith review of: GOSU: Retrieval-Augmented Generation with Global-Level Optimized Semantic Unit-Centric Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WZEFTHBY}},
  note         = {Machine review of arXiv:2509.00449}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Building upon the standard graph-based Retrieval-Augmented Generation (RAG), the introduction of heterogeneous graphs and hypergraphs aims to enrich retrieval and generation by leveraging the relationships between multiple entities through the concept of semantic units (SUs). But this also raises a key issue: The extraction of high-level SUs limited to local text chunks is prone to ambiguity, complex coupling, and increased retrieval overhead due to the lack of global knowledge or the neglect of fine-grained relationships. To address these issues, we propose GOSU, a semantic unit-centric RAG framework that efficiently performs global disambiguation and utilizes SUs to capture interconnections between different nodes across the global context. In the graph construction phase, GOSU performs global merging on the pre-extracted SUs from local text chunks and guides entity and relationship extraction, reducing the difficulty of coreference resolution while uncovering global semantic objects across text chunks. In the retrieval and generation phase, we introduce hierarchical keyword extraction and semantic unit completion. The former uncovers the fine-grained binary relationships overlooked by the latter, while the latter compensates for the coarse-grained n-ary relationships missing from the former. Evaluation across multiple tasks demonstrates that GOSU outperforms the baseline RAG methods in terms of generation quality.

Figures

Figures reproduced from arXiv: 2509.00449 by Bingbing Wang, Huafei Deng, Ke Liu, Li Zhang, Xuecheng Zou, Yu Tang.

Figure 1
Figure 1. Figure 1: Comparison of graph-based RAG structures. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An overview of GOSU, which extracts semantic units from the corpus and applies global-level optimization to [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Tokens comparision and cost comparison of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

89 extracted references · 13 canonical work pages · 9 internal anchors

  1. [1]

    Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. 2019. https://arxiv.org/abs/1911.10470 Learning to retrieve reasoning paths over Wikipedia graph for question answering . Preprint, arXiv:1911.10470. Preprint, arXiv:1911.10470

    Pith/arXiv arXiv 2019

  2. [2]

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2023. https://arxiv.org/abs/2310.11511 Self-RAG : Learning to retrieve, generate, and critique through Self-Reflection . Preprint, arXiv:2310.11511. Preprint, arXiv:2310.11511

    Pith/arXiv arXiv 2023

  3. [3]

    Zechen Bai, Pichao Wang, Tianjun Xiao, Tong He, Zongbo Han, Zheng Zhang, and Mike Zheng Shou. 2024. https://arxiv.org/abs/2404.18930 Hallucination of Multimodal Large Language Models : A survey . Preprint, arXiv:2404.18930. Preprint, arXiv:2404.18930

    Pith/arXiv arXiv 2024

  4. [4]

    Yejin Bang, Ziwei Ji, Alan Schelten, Anthony Hartshorn, Tara Fowler, Cheng Zhang, Nicola Cancedda, and Pascale Fung. 2025. https://arxiv.org/abs/2504.17550 HalluLens : LLM hallucination benchmark . Preprint, arXiv:2504.17550. Preprint, arXiv:2504.17550

    Pith/arXiv arXiv 2025

  5. [5]

    Joel Barmettler, Abraham Bernstein, and Luca Rossetto. 2025. https://arxiv.org/abs/2504.07624 ConceptFormer : Towards efficient use of Knowledge-Graph embeddings in Large Language Models . Preprint, arXiv:2504.07624. Preprint, arXiv:2504.07624

    Pith/arXiv arXiv 2025

  6. [6]

    Dominique Brunato, Felice Dell’Orletta, Irene Dini, and Andrea Amelio Ravelli. 2023. https://aclanthology.org/2023.findings-acl.94/ Coherent or not? stressing a neural language model for discourse coherence in multiple languages . In Findings of the Association for Computational Linguistics: ACL 2023 , pages 10690--10700, Toronto, Canada. Association for ...

    2023

  7. [7]

    Weijie Chen, Ting Bai, Jinbo Su, Jian Luan, Wei Liu, and Chuan Shi. 2024. https://arxiv.org/abs/2412.05547 KG-Retriever : Efficient knowledge indexing for Retrieval-Augmented Large Language Models . Preprint, arXiv:2412.05547. Preprint, arXiv:2412.05547

    Pith/arXiv arXiv 2024

  8. [8]

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, and Jonathan Larson. 2024. https://arxiv.org/abs/2404.16130 From local to global: A graph RAG approach to Query-Focused Summarization . Preprint, arXiv:2404.16130. Preprint, arXiv:2404.16130

    Pith/arXiv arXiv 2024

  9. [9]

    Shahul Es, Jithin James, Luis Espinosa Anke, and Steven Schockaert. 2024. https://doi.org/10.18653/v1/2024.eacl-demo.16 RAGAs : Automated evaluation of Retrieval Augmented Generation . In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations, pages 150--158, St. Julian's, Malta. ...

    doi:10.18653/v1/2024.eacl-demo.16 2024

  10. [10]

    Wenqi Fan, Yujuan Ding, Liangbo Ning, Shijie Wang, Hengyun Li, Dawei Yin, Tat-Seng Chua, and Qing Li. 2024. https://doi.org/10.1145/3637528.3671470 A survey on RAG meeting LLMs : Towards Retrieval-Augmented large language models . In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 6491--6501, Barcelona, Spain. ...

    arXiv 2024

  11. [11]

    Luyu Gao, Xueguang Ma, Jimmy Lin, and Jamie Callan. 2023 a . https://doi.org/10.18653/v1/2023.acl-long.99 Precise zero-shot dense retrieval without relevance labels . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1762--1777, Toronto, Canada. Association for Computational Linguistics

    doi:10.18653/v1/2023.acl-long.99 2023

  12. [12]

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Meng Wang, and Haofen Wang. 2023 b . https://arxiv.org/abs/2312.10997 Retrieval-Augmented Generation for Large Language Models : A survey . Preprint, arXiv:2312.10997. Preprint, arXiv:2312.10997

    Pith/arXiv arXiv 2023

  13. [13]

    Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang. 2024. https://arxiv.org/abs/2410.05779 LightRAG : Simple and fast Retrieval-Augmented Generation . Preprint, arXiv:2410.05779. Preprint, arXiv:2410.05779

    Pith/arXiv arXiv 2024

  14. [14]

    Rossi, Subhabrata Mukherjee, Xianfeng Tang, Qi He, Zhigang Hua, Bo Long, Tong Zhao, Neil Shah, Amin Javari, Yinglong Xia, and Jiliang Tang

    Haoyu Han, Yu Wang, Harry Shomer, Kai Guo, Jiayuan Ding, Yongjia Lei, Mahantesh Halappanavar, Ryan A. Rossi, Subhabrata Mukherjee, Xianfeng Tang, Qi He, Zhigang Hua, Bo Long, Tong Zhao, Neil Shah, Amin Javari, Yinglong Xia, and Jiliang Tang. 2024. https://arxiv.org/abs/2501.00309 Retrieval-augmented generation with graphs ( GraphRAG ) . Preprint, arXiv:25...

    Pith/arXiv arXiv 2024

  15. [15]

    Chawla, Thomas Laurent, Yann LeCun, Xavier Bresson, and Bryan Hooi

    Xiaoxin He, Yijun Tian, Yifei Sun, Nitesh V. Chawla, Thomas Laurent, Yann LeCun, Xavier Bresson, and Bryan Hooi. 2024. https://arxiv.org/abs/2402.07630 G-retriever: Retrieval-augmented generation for textual graph understanding and question answering . In Advances in Neural Information Processing Systems, volume 37, pages 132876--132907. Curran Associates, Inc

    Pith/arXiv arXiv 2024

  16. [16]

    Wentao Hu, Wengyu Zhang, Yiyang Jiang, Chen Jason Zhang, Xiaoyong Wei, and Qing Li. 2025. https://arxiv.org/abs/2505.18581 Removal of Hallucination on Hallucination : Debate-Augmented RAG . Preprint, arXiv:2505.18581. Preprint, arXiv:2505.18581

    Pith/arXiv arXiv 2025

  17. [17]

    Haoyu Huang, Yongfeng Huang, Junjie Yang, Zhenyu Pan, Yongqiang Chen, Kaili Ma, Hongzhi Chen, and James Cheng. 2025. https://arxiv.org/abs/2503.10150 HiRAG : Retrieval-Augmented Generation with Hierarchical Knowledge . Preprint, arXiv:2503.10150. Preprint, arXiv:2503.10150

    arXiv 2025

  18. [18]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2023. https://arxiv.org/abs/2311.05232 A survey on hallucination in Large Language Models : Principles, taxonomy, challenges, and open questions . Preprint, arXiv:2311.05232. Preprint, arXiv:2311.05232

    Pith/arXiv arXiv 2023

  19. [19]

    Gautier Izacard and Edouard Grave. 2020. https://arxiv.org/abs/2007.01282 Leveraging Passage Retrieval with Generative Models for Open-Domain Question Answering . Preprint, arXiv:2007.01282. Preprint, arXiv:2007.01282

    Pith/arXiv arXiv 2020

  20. [20]

    Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. 2022. https://arxiv.org/abs/2208.03299 ATLAS : Few-shot learning with Retrieval Augmented language models . Preprint, arXiv:2208.03299. Preprint, arXiv:2208.03299

    Pith/arXiv arXiv 2022

  21. [21]

    Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Yejin Bang, Delong Chen, Wenliang Dai, Ho Shu Chan, Andrea Madotto, and Pascale Fung. 2022. https://arxiv.org/abs/2202.03629 Survey of Hallucination in Natural Language Generation . Preprint, arXiv:2202.03629. Preprint, arXiv:2202.03629

    Pith/arXiv arXiv 2022

  22. [22]

    Xinke Jiang, Rihong Qiu, Yongxin Xu, Wentao Zhang, Yichen Zhu, Ruizhe Zhang, Yuchen Fang, Xu Chu, Junfeng Zhao, and Yasha Wang. 2024. https://arxiv.org/abs/2410.23855 RAGraph : A general Retrieval-Augmented graph learning framework . In Advances in Neural Information Processing Systems, volume 37, pages 29948--29985. Curran Associates, Inc

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  23. [23]

    Zhengbao Jiang, Frank Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan, and Graham Neubig. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.495 Active retrieval augmented generation . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 7969--7992, Singapore. Association for Compu...

    doi:10.18653/v1/2023.emnlp-main.495 2023

  24. [24]

    Yihan Jiao, Zhehao Tan, Dan Yang, Duolin Sun, Jie Feng, Yue Shen, Jian Wang, and Peng Wei. 2025. https://arxiv.org/abs/2507.05714 HIRAG : Hierarchical-thought instruction-tuning Retrieval-Augmented Generation . Preprint, arXiv:2507.05714. Preprint, arXiv:2507.05714

    Pith/arXiv arXiv 2025

  25. [25]

    Bernal Jim \'e nez Guti \'e rrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, and Yu Su. 2024. https://arxiv.org/abs/2405.14831 Hipporag: Neurobiologically inspired long-term memory for large language models . In Advances in Neural Information Processing Systems, volume 37, pages 59532--59569. Curran Associates, Inc

    Pith/arXiv arXiv 2024

  26. [26]

    Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. https://aclanthology.org/2020.emnlp-main.550/ Dense passage retrieval for open-domain question answering . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing ( EMNLP ) , pages 6769--6781, Online. As...

    2020

  27. [27]

    Gangwoo Kim, Sungdong Kim, Byeongguk Jeon, Joonsuk Park, and Jaewoo Kang. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.63 Tree of clarifications: Answering ambiguous questions with retrieval-augmented large language models . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 996--1009, Singapore. Associat...

    doi:10.18653/v1/2023.emnlp-main.63 2023

  28. [28]

    u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K \"u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \"a schel, Sebastian Riedel, and Douwe Kiela. 2020. https://arxiv.org/abs/2005.11401 Retrieval-Augmented Generation for knowledge-intensive NLP tasks . Preprint, arXiv:2005.11401. Preprint, arXiv:2005.11401

    Pith/arXiv arXiv 2020

  29. [29]

    Lei Liang, Mengshu Sun, Zhengke Gui, Zhongshu Zhu, Zhouyu Jiang, Ling Zhong, Yuan Qu, Peilong Zhao, Zhongpu Bo, Jin Yang, Huaidong Xiong, Lin Yuan, Jun Xu, Zaoyang Wang, Zhiqiang Zhang, Wen Zhang, Huajun Chen, Wenguang Chen, and Jun Zhou. 2025. https://doi.org/10.1145/3701716.3715240 KAG : Boosting LLMs in professional domains via Knowledge Augmented Gene...

    arXiv 2025

  30. [30]

    Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

    Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2023. https://arxiv.org/abs/2307.03172 Lost in the middle: How Language Models use long contexts . Preprint, arXiv:2307.03172. Preprint, arXiv:2307.03172

    Pith/arXiv arXiv 2023

  31. [31]

    Zuhong Liu, Charles-Elie Simon, and Fabien Caspani. 2025. https://doi.org/10.1007/978-3-031-88714-7_33 Passage segmentation of documents for extractive question answering . In European Conference on Information Retrieval, pages 345--352, Cham, Switzerland. Springer Nature Switzerland

    work page doi:10.1007/978-3-031-88714-7_33 2025

  32. [32]

    Haoran Luo, Haihong E, Guanting Chen, Yandan Zheng, Xiaobao Wu, Yikai Guo, Qika Lin, Yu Feng, Zemin Kuang, Meina Song, Yifan Zhu, and Anh Tuan Luu. 2025 a . https://arxiv.org/abs/2503.21322 HyperGraphRAG : Retrieval-Augmented Generation via Hypergraph-Structured Knowledge Representation . Preprint, arXiv:2503.21322. Preprint, arXiv:2503.21322

    arXiv 2025

  33. [33]

    Linhao Luo, Yuan-Fang Li, Gholamreza Haffari, and Shirui Pan. 2023. https://arxiv.org/abs/2310.01061 Reasoning on Graphs : Faithful and interpretable Large Language Model reasoning . Preprint, arXiv:2310.01061. Preprint, arXiv:2310.01061

    Pith/arXiv arXiv 2023

  34. [34]

    Linhao Luo, Zicheng Zhao, Gholamreza Haffari, Dinh Phung, Chen Gong, and Shirui Pan. 2025 b . https://arxiv.org/abs/2502.01113 GFM-RAG : Graph foundation model for Retrieval-Augmented Generation . Preprint, arXiv:2502.01113. Preprint, arXiv:2502.01113

    arXiv 2025

  35. [35]

    Yougang Lyu, Lingyong Yan, Shuaiqiang Wang, Haibo Shi, Dawei Yin, Pengjie Ren, Zhumin Chen, Maarten de Rijke, and Zhaochun Ren. 2024. https://arxiv.org/abs/2402.11176 KnowTuning : Knowledge-aware fine-tuning for Large Language Models . Preprint, arXiv:2402.11176. Preprint, arXiv:2402.11176

    Pith/arXiv arXiv 2024

  36. [36]

    Chuangtao Ma, Yongrui Chen, Tianxing Wu, Arijit Khan, and Haofen Wang. 2025. https://arxiv.org/abs/2505.20099 Large Language Models meet Knowledge Graphs for Question Answering : Synthesis and opportunities . Preprint, arXiv:2505.20099. Preprint, arXiv:2505.20099

    arXiv 2025

  37. [37]

    Kaixin Ma, Hao Cheng, Xiaodong Liu, Eric Nyberg, and Jianfeng Gao. 2022. https://arxiv.org/abs/2210.12338 Open-domain question answering via chain of reasoning over heterogeneous knowledge . Preprint, arXiv:2210.12338. Preprint, arXiv:2210.12338

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  38. [38]

    Alex Mallen, Akari Asai, Victor Zhong, Rajarshi Das, Daniel Khashabi, and Hannaneh Hajishirzi. 2022. https://arxiv.org/abs/2212.10511 When not to trust Language Models : Investigating effectiveness of parametric and non-parametric memories . Preprint, arXiv:2212.10511. Preprint, arXiv:2212.10511

    Pith/arXiv arXiv 2022

  39. [39]

    Costas Mavromatis and George Karypis. 2024. https://arxiv.org/abs/2405.20139 GNN-RAG : Graph neural retrieval for Large Language Model reasoning . Preprint, arXiv:2405.20139. Preprint, arXiv:2405.20139

    Pith/arXiv arXiv 2024

  40. [40]

    McCarthy, Rosa Maria Bruno, John W

    Cian P. McCarthy, Rosa Maria Bruno, John W. McEvoy, and Rhian M. Touyz. 2025. https://doi.org/10.1093/ehjcvp/pvae084 2024 ESC guidelines for the management of elevated blood pressure and hypertension: What is new in pharmacotherapy? European Heart Journal - Cardiovascular Pharmacotherapy, 11(1):7--9

    work page doi:10.1093/ehjcvp/pvae084 2025

  41. [41]

    Lingrui Mei, Jiayu Yao, Yuyao Ge, Yiwei Wang, Baolong Bi, Yujun Cai, Jiazhi Liu, Mingyu Li, Zhong-Zhi Li, Duzhen Zhang, Chenlin Zhou, Jiayi Mao, Tianze Xia, Jiafeng Guo, and Shenghua Liu. 2025. https://arxiv.org/abs/2507.13334 A survey of Context Engineering for Large Language Models . Preprint, arXiv:2507.13334. Preprint, arXiv:2507.13334

    Pith/arXiv arXiv 2025

  42. [42]

    Norman, and Mariya Toneva

    Sebastian Michelmann, Manoj Kumar, Kenneth A. Norman, and Mariya Toneva. 2025. Large language models can segment narrative events similarly to humans. Behavior Research Methods, 57(1):39

    2025

  43. [43]

    Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Wei Koh, Mohit Iyyer, Luke Zettlemoyer, and Hannaneh Hajishirzi. 2023. https://arxiv.org/abs/2305.14251 FActScore : Fine-grained atomic evaluation of factual precision in long-form text generation . Preprint, arXiv:2305.14251. Preprint, arXiv:2305.14251

    Pith/arXiv arXiv 2023

  44. [44]

    Sewon Min, Kenton Lee, Ming-Wei Chang, Kristina Toutanova, and Hannaneh Hajishirzi. 2021. https://arxiv.org/abs/2104.08445 Joint passage ranking for diverse multi-answer retrieval . Preprint, arXiv:2104.08445. Preprint, arXiv:2104.08445

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  45. [45]

    Tongke Ni, Yang Fan, Junru Zhou, Xiangping Wu, and Qingcai Chen. 2025. https://arxiv.org/abs/2503.23671 CrossFormer : Cross-segment semantic fusion for document segmentation . Preprint, arXiv:2503.23671. Preprint, arXiv:2503.23671

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  46. [46]

    Cheng Niu, Yuanhao Wu, Juno Zhu, Siliang Xu, Kashun Shum, Randy Zhong, Juntong Song, and Tong Zhang. 2024. https://arxiv.org/abs/2401.00396 RAGTruth : A hallucination corpus for developing trustworthy Retrieval-Augmented Language Models . Preprint, arXiv:2401.00396. Preprint, arXiv:2401.00396

    Pith/arXiv arXiv 2024

  47. [47]

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, Irwan Bello, Jake Berdine, Gabriel Bernadett-Shapiro, Christopher Berner...

    Pith/arXiv arXiv 2023

  48. [48]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. https://arxiv.org/abs/2203.02155 Training language models t...

    Pith/arXiv arXiv 2022

  49. [49]

    Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. 2023. https://arxiv.org/abs/2306.08302 Unifying Large Language Models and Knowledge Graphs : A roadmap . Preprint, arXiv:2306.08302. Preprint, arXiv:2306.08302

    Pith/arXiv arXiv 2023

  50. [50]

    Graph Elicitation for Guiding Multi-Step Reasoning in Large Language Models

    Jinyoung Park, Ameen Patel, Omar Zia Khan, Hyunwoo J. Kim, and Joo-Kyung Kim. 2023. https://arxiv.org/abs/2311.09762 Graph elicitation for guiding multi-step reasoning in large language models . Preprint, arXiv:2311.09762. Preprint, arXiv:2311.09762

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  51. [51]

    Boci Peng, Yun Zhu, Yongchao Liu, Xiaohe Bo, Haizhou Shi, Chuntao Hong, Yan Zhang, and Siliang Tang. 2024. https://arxiv.org/abs/2408.08921 Graph Retrieval-Augmented Generation : A survey . Preprint, arXiv:2408.08921. Preprint, arXiv:2408.08921

    Pith/arXiv arXiv 2024

  52. [52]

    Pouya Pezeshkpour and Estevam Hruschka. 2023. https://arxiv.org/abs/2308.11483 Large Language Models sensitivity to the order of options in multiple-choice questions . Preprint, arXiv:2308.11483. Preprint, arXiv:2308.11483

    Pith/arXiv arXiv 2023

  53. [53]

    Hongjin Qian, Zheng Liu, Peitian Zhang, Kelong Mao, Defu Lian, Zhicheng Dou, and Tiejun Huang. 2024. https://arxiv.org/abs/2409.05591 MemoRAG : Boosting long context processing with global memory-enhanced Retrieval Augmentation . Preprint, arXiv:2409.05591. Preprint, arXiv:2409.05591

    Pith/arXiv arXiv 2024

  54. [54]

    Ladislav Ramp \'a s ek, Mikhail Galkin, Vijay Prakash Dwivedi, Anh Tuan Luu, Guy Wolf, and Dominique Beaini. 2022. https://arxiv.org/abs/2205.12454 Recipe for a general, powerful, scalable Graph Transformer . In Advances in Neural Information Processing Systems, volume 35, pages 14501--14515. Curran Associates, Inc

    Pith/arXiv arXiv 2022

  55. [55]

    Alireza Salemi and Hamed Zamani. 2024. https://doi.org/10.1145/3626772.3657957 Evaluating retrieval quality in Retrieval-Augmented Generation . In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages 2395--2400, Washington, DC, USA. Association for Computing Machinery

    arXiv 2024

  56. [56]

    Diego Sanmartin. 2024. https://arxiv.org/abs/2405.12035 Kg-rag: Bridging the gap between knowledge and creativity . Preprint, arXiv:2405.12035. Preprint, arXiv:2405.12035

    Pith/arXiv arXiv 2024

  57. [57]

    Zhihong Shao, Yeyun Gong, Yelong Shen, Minlie Huang, Nan Duan, and Weizhu Chen. 2023. https://arxiv.org/abs/2305.15294 Enhancing Retrieval-Augmented Large Language Models with Iterative Retrieval-Generation Synergy . Preprint, arXiv:2305.15294. Preprint, arXiv:2305.15294

    Pith/arXiv arXiv 2023

  58. [58]

    Zhili Shen, Chenxin Diao, Pavlos Vougiouklis, Pascual Merita, Shriram Piramanayagam, Enting Chen, Damien Graux, Andre Melo, Ruofei Lai, Zeren Jiang, Zhongyang Li, Ye Qi, Yang Ren, Dandan Tu, and Jeff Z. Pan. 2024. https://arxiv.org/abs/2412.18431 GeAR : Graph-enhanced agent for Retrieval-Augmented Generation . Preprint, arXiv:2412.18431. Preprint, arXiv:2...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  59. [59]

    Robik Shrestha, Yang Zou, Qiuyu Chen, Zhiheng Li, Yusheng Xie, and Siqi Deng. 2024. https://openaccess.thecvf.com/content/CVPR2024/html/Shrestha_FairRAG_Fair_Human_Generation_via_Fair_Retrieval_Augmentation_CVPR_2024_paper.html FairRAG : Fair human generation via fair retrieval augmentation . In Proceedings of the IEEE/CVF Conference on Computer Vision an...

    2024

  60. [60]

    Valentin Six, Evan Dufraisse, and Ga \"e l de Chalendar. 2025. https://arxiv.org/abs/2506.13380 Decompositional reasoning for graph retrieval with Large Language Models . Preprint, arXiv:2506.13380. Preprint, arXiv:2506.13380

    work page arXiv 2025

  61. [61]

    Weihang Su, Yichen Tang, Qingyao Ai, Zhijing Wu, and Yiqun Liu. 2024. https://arxiv.org/abs/2403.10081 DRAGIN : Dynamic Retrieval-Augmented Generation based on the Information Needs of Large Language Models . Preprint, arXiv:2403.10081. Preprint, arXiv:2403.10081

    Pith/arXiv arXiv 2024

  62. [62]

    Viju Sudhi, Sinchana Ramakanth Bhat, Max Rudat, and Roman Teucher. 2024. https://doi.org/10.1145/3626772.3657660 RAG-Ex : A generic framework for explaining Retrieval Augmented Generation . In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages 2776--2780, Washington, DC, USA. Association...

    arXiv 2024

  63. [63]

    Haitian Sun, Tania Bedrax-Weiss, and William W. Cohen. 2019. https://arxiv.org/abs/1904.09537 PullNet : Open-domain question answering with iterative retrieval on knowledge bases and text . Preprint, arXiv:1904.09537. Preprint, arXiv:1904.09537

    Pith/arXiv arXiv 2019

  64. [64]

    Chawla, and Panpan Xu

    Yijun Tian, Huan Song, Zichen Wang, Haozhu Wang, Ziqing Hu, Fang Wang, Nitesh V. Chawla, and Panpan Xu. 2024. https://doi.org/10.1609/aaai.v38i17.29875 Graph neural prompting with Large Language Models . Proceedings of the AAAI Conference on Artificial Intelligence , 38(17):19080--19088

    doi:10.1609/aaai.v38i17.29875 2024

  65. [65]

    S. M. Towhidul Islam Tonmoy, S. M. Mehedi Zaman, Vinija Jain, Anku Rani, Vipula Rawte, Aman Chadha, and Amitava Das. 2024. https://arxiv.org/abs/2401.01313 A comprehensive survey of Hallucination mitigation techniques in Large Language Models . Preprint, arXiv:2401.01313. Preprint, arXiv:2401.01313

    Pith/arXiv arXiv 2024

  66. [66]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller, Cynthia Gao, Vedanuj Goswami, Naman Goyal, Anthony Harts...

    Pith/arXiv arXiv 2023

  67. [67]

    Milena Trajanoska, Riste Stojanov, and Dimitar Trajanov. 2023. https://arxiv.org/abs/2305.04676 Enhancing knowledge graph construction using large language models . Preprint, arXiv:2305.04676. Preprint, arXiv:2305.04676

    Pith/arXiv arXiv 2023

  68. [68]

    Harsh Trivedi, Niranjan Balasubramanian, Tushar Khot, and Ashish Sabharwal. 2022. https://arxiv.org/abs/2212.10509 Interleaving retrieval with Chain-of-Thought reasoning for Knowledge-Intensive multi-step questions . Preprint, arXiv:2212.10509. Preprint, arXiv:2212.10509

    Pith/arXiv arXiv 2022

  69. [69]

    Shangqing Tu, Yuanchun Wang, Jifan Yu, Yuyang Xie, Yaran Shi, Xiaozhi Wang, Jing Zhang, Lei Hou, and Juanzi Li. 2024. https://doi.org/10.1145/3637528.3671564 R-Eval : A unified toolkit for evaluating domain knowledge of Retrieval Augmented Large Language Models . In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pag...

    arXiv 2024

  70. [70]

    Jinyu Wang, Jingjing Fu, Rui Wang, Lei Song, and Jiang Bian. 2025 a . https://arxiv.org/abs/2501.11551 PIKE-RAG : Specialized knowledge and rationale augmented generation . Preprint, arXiv:2501.11551. Preprint, arXiv:2501.11551

    Pith/arXiv arXiv 2025

  71. [71]

    Shu Wang, Yixiang Fang, Yingli Zhou, Xilin Liu, and Yuchi Ma. 2025 b . https://arxiv.org/abs/2502.09891 ArchRAG : Attributed community-based hierarchical Retrieval-Augmented Generation . Preprint, arXiv:2502.09891. Preprint, arXiv:2502.09891

    Pith/arXiv arXiv 2025

  72. [72]

    Xiaohua Wang, Zhenghua Wang, Xuan Gao, Feiran Zhang, Yixin Wu, Zhibo Xu, Tianyuan Shi, Zhengyuan Wang, Shizheng Li, Qi Qian, Ruicheng Yin, Changze Lv, Xiaoqing Zheng, and Xuanjing Huang. 2024 a . https://arxiv.org/abs/2407.01219 Searching for best practices in Retrieval-Augmented Generation . Preprint, arXiv:2407.01219. Preprint, arXiv:2407.01219

    Pith/arXiv arXiv 2024

  73. [73]

    Rossi, Alexa Siu, Ruiyi Zhang, and Tyler Derr

    Yu Wang, Nedim Lipka, Ryan A. Rossi, Alexa Siu, Ruiyi Zhang, and Tyler Derr. 2024 b . https://doi.org/10.1609/aaai.v38i17.29889 Knowledge graph prompting for Multi-Document Question Answering . Proceedings of the AAAI Conference on Artificial Intelligence, 38(17):19206--19214

    work page doi:10.1609/aaai.v38i17.29889 2024

  74. [74]

    Zhepei Wei, Wei-Lin Chen, and Yu Meng. 2024. https://arxiv.org/abs/2406.13629 InstructRAG : Instructing Retrieval-Augmented Generation via self-synthesized rationales . Preprint, arXiv:2406.13629. Preprint, arXiv:2406.13629

    Pith/arXiv arXiv 2024

  75. [75]

    Jianfeng Wen, Jianxin Li, Yongyi Mao, Shini Chen, and Richong Zhang. 2016. https://arxiv.org/abs/1604.08642 On the representation and embedding of knowledge bases beyond binary relations . Preprint, arXiv:1604.08642. Preprint, arXiv:1604.08642

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  76. [76]

    Yilin Wen, Zifeng Wang, and Jimeng Sun. 2023. https://arxiv.org/abs/2308.09729 MindMap : Knowledge graph prompting sparks Graph of Thoughts in Large Language Models . Preprint, arXiv:2308.09729. Preprint, arXiv:2308.09729

    Pith/arXiv arXiv 2023

  77. [77]

    Zhishang Xiang, Chuanjie Wu, Qinggang Zhang, Shengyuan Chen, Zijin Hong, Xiao Huang, and Jinsong Su. 2025. https://arxiv.org/abs/2506.05690 When to use Graphs in RAG : A comprehensive analysis for Graph Retrieval-Augmented Generation . Preprint, arXiv:2506.05690. Preprint, arXiv:2506.05690

    arXiv 2025

  78. [78]

    Derong Xu, Pengyue Jia, Xiaopeng Li, Yingyi Zhang, Maolin Wang, Qidong Liu, Xiangyu Zhao, Yichao Wang, Huifeng Guo, Ruiming Tang, Enhong Chen, and Tong Xu. 2025 a . https://arxiv.org/abs/2505.16237 Align-GRAG : Reasoning-guided dual alignment for Graph Retrieval-Augmented Generation . Preprint, arXiv:2505.16237. Preprint, arXiv:2505.16237

    arXiv 2025

  79. [79]

    Tianyang Xu, Haojie Zheng, Chengze Li, Haoxiang Chen, Yixin Liu, Ruoxi Chen, and Lichao Sun. 2025 b . https://arxiv.org/abs/2504.11544 NodeRAG : Structuring graph-based RAG with heterogeneous nodes . Preprint, arXiv:2504.11544. Preprint, arXiv:2504.11544

    Pith/arXiv arXiv 2025

  80. [80]

    Hai Yu, Chong Deng, Qinglin Zhang, Jiaqing Liu, Qian Chen, and Wen Wang. 2023. https://arxiv.org/abs/2310.11772 Improving long document topic segmentation models with enhanced coherence modeling . Preprint, arXiv:2310.11772. Preprint, arXiv:2310.11772

    work page internal anchor Pith review Pith/arXiv arXiv 2023

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.