Pith. sign in

REVIEW 5 major objections 6 minor 42 references

MoLoRAG: Bootstrapping Document Understanding via Multi-modal Logic-aware Retrieval

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Graph-walking VLM finds evidence pages that semantic search misses

desk verdict A useful training-free retrieval module for multi-page DocQA, but the fine-tuned variant's gains are compromised by training/eval overlap. read the letter →

arxiv 2509.07666 v1 pith:3IYXJQDM submitted 2025-09-06 cs.CL cs.IR

Xixi Wu , Yanchao Tan , Nan Hou , Ruiyang Zhang , Hong Cheng This is my paper · ORCID
classification cs.CLcs.IR
keywords documentquestionansweringmulti-modalretrievalpagegraphlogicalrelevancetraversalvision-languagemodelretrieval-augmentedgenerationmulti-pagedocuments
verification ladder T0 review T1 audit T2 compute T3 formal

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 argues that document question answering over long, image-heavy documents fails when retrieval looks only at semantic similarity between the query and each page. Its proposed answer, MoLoRAG, builds a graph of pages whose embeddings are similar and lets a lightweight vision-language model walk that graph, giving each visited page a logical-relevance score from 1 to 5. The final ranking combines semantic and logical scores, and the top pages are passed to any LVLM for answering. On four DocQA datasets the paper reports that this consistently beats semantic-only retrievers, with average retrieval improvements of 9.94 percent on MMLongBench and 7.16 percent on LongDocURL, and that retrieval gains carry through to answer accuracy. The underlying point is that reasoning questions often need evidence pages that are logically connected to the question rather than merely keyword-similar.

What carries the argument

The load-bearing mechanism is the page graph G(V,E) constructed from ColPali page embeddings, with an edge when the inner product exceeds threshold θ. Graph traversal, seeded by the top-w semantic pages and limited by a hop count, lets the retrieval engine reach pages that are not semantically similar to the query but are connected to similar pages; the VLM's 1-5 logical-relevance score, combined with the semantic score, decides which pages survive the re-ranking. The graph supplies the candidate paths, the VLM supplies the new logical signal, and the two hyperparameters w and n_hop keep the search from visiting every page.

What would settle it

Construct a split of MMLongBench and LongDocURL where every ground-truth evidence page is more than n_hop=4 graph hops away from the top-3 semantic pages at θ=0.4. If MoLoRAG still retrieves those evidence pages, graph traversal is not what supplies them; if it fails, the load-bearing assumption is confirmed. A complementary check is to measure whether the VLM's 1-5 score discriminates ground-truth evidence pages from non-evidence pages at better than chance.

Watch

Extended reading notes

Core claim

MoLoRAG's central claim is that a page graph plus a VLM-assigned logical relevance score retrieves the evidence pages needed to answer multi-page document questions, where semantic retrieval alone returns pages containing query keywords but lacking the information. The method encodes each page with ColPali, adds an edge when embedding similarity exceeds a threshold, initializes an exploration set from the top semantically similar pages, and then iteratively scores neighbors with a VLM prompt (1-5), combines that score with the semantic score, and expands the set until the hop limit is reached. All visited pages are re-ranked by the combined score, and the top-K snapshots are sent to an LVLM. The paper reports that this consistently outperforms baselines, with an average improvement of 9.94 percent on MMLongBench and 7.16 percent on LongDocURL in retrieval metrics, and that a fine-tuned variant MoLoRAG+ improves further by making the logical-relevance judgments more accurate.

Load-bearing premise

The method assumes that the page graph, built from embedding similarity above a fixed threshold, links semantically similar pages to the logically relevant evidence pages within a few hops, and that a VLM's one-to-five relevance score recognizes those evidence pages when it sees them.

Editorial extensions

If this is right

  • MoLoRAG is compatible with arbitrary LVLMs: once retrieval is done, only the top-K page snapshots are passed in, so single-image models can be used on multi-page documents.
  • Retrieval gains are largest on questions whose evidence page is logically, not lexically, related to the query; the paper's case studies show such questions are precisely where semantic-only baselines fail.
  • The training-free variant is deployable without extra data, while the fine-tuned MoLoRAG+ variant provides a further accuracy boost by improving logical relevance scoring.
  • On long documents the traversal queries fewer than half the pages on average, so the method scales to documents too large for direct LVLM input.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: if logical relevance is the bottleneck, then the page graph's edge construction deserves at least as much attention as the scorer; replacing the similarity threshold with structural edges from tables of contents, hyperlinks, or repeated entities should raise recall without retraining.
  • The discrete 1-5 VLM score creates ties, and the paper itself notes that equal scores make ranking hard; a continuous regression head or pairwise preference model trained on the released triplets is a natural extension.
  • A testable extension is threshold sensitivity: recall as a function of θ should show a knee, and per-document adaptive thresholds could preserve logical edges at the cost of more VLM calls.
  • The paper's limitation section notes it targets closed-domain documents; carrying the same idea to open-domain corpora would require cross-document graphs, which the current traversal does not model.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes MoLoRAG, a two-stage retrieval-augmented framework for multi-page document QA. In the retrieval stage, a page graph is built from ColPali embedding similarities; a lightweight VLM assigns a 1–5 'logical relevance' score to each visited page while traversing the graph, and these scores are combined with semantic scores to re-rank pages. The top-K pages are then passed to an arbitrary LVLM for answer generation. A fine-tuned variant, MoLoRAG+, is trained on GPT-4o-generated triplets to improve logical-relevance scoring. Experiments on MMLongBench, LongDocURL, PaperTab, and FetaTab report consistent QA and retrieval gains over M3DocRAG, MDocAgent, and direct LVLM inference, with additional ablations for the logical-only variant and the full-traversal variant.

Significance. If the claims hold, MoLoRAG is a useful contribution: it extends graph-RAG ideas to multimodal page retrieval, offers a training-free variant that is easy to deploy, and releases code, model weights, and a training dataset. The central idea of combining semantic similarity with VLM-assigned logical relevance is plausible and the paper provides a concrete instantiation with ablations. The empirical scope is broad (four datasets, multiple LVLMs) and the paper includes useful analyses of efficiency and fine-grained modality performance. However, the significance is currently conditional on resolving the train/evaluation overlap for MoLoRAG+, the truncated direct baseline, and the lack of uncertainty quantification.

major comments (5)
  1. [Appendix D.2 and Table 2] The fine-tuned variant MoLoRAG+ is trained on 3,519 page snapshots 'randomly selected from MMLongBench and LongDocURL' and then evaluated on the same two benchmarks, but the paper does not describe any document-level or page-level split between training and evaluation. If the sampled snapshots come from the same documents or pages as the evaluation queries, the reported gains (e.g., MoLoRAG+ top-3 Recall 68.87 versus MoLoRAG 67.22 on MMLongBench, and top-1 Recall 51.32 versus 45.46) may reflect memorized visual content rather than improved logical-relevance reasoning. This directly affects the strongest rows of Table 3 and the corresponding QA improvements in Table 2. The authors should either use a document-disjoint split for fine-tuning and evaluation, or explicitly demonstrate that no overlap exists between the training snapshots and the evaluation pages.
  2. [Tables 2 and 3] The 'LVLM Direct' baseline truncates documents to the first 30 pages, while LongDocURL has an average of 85.6 pages per document and the case studies in Figures 5 and 8 place evidence pages at indices 49 and 51. Truncation therefore handicaps the direct baseline on long documents, making the 'Direct' comparisons in Table 2 and the claimed average improvements over direct inference difficult to interpret. The authors should compare against direct inference that processes the full document where the LVLM context allows it, or clearly report the percentage of LongDocURL questions whose evidence pages fall outside the 30-page window.
  3. [Appendix D.6 and Table 8] All reported numbers are single-run point estimates without error bars, multiple seeds, or significance tests. Some differences that support the paper's claims are small—for example Qwen2.5-VL-3B on MMLongBench: MoLoRAG 32.11 versus MoLoRAG+ 32.47 in Table 2, and LongDocURL top-3 Recall 70.04 versus 68.92 in Table 3. Without variance estimates, it is not possible to tell whether these differences are meaningful. The authors should report at least three independent runs with standard deviations, or a paired significance test, for the main retrieval and QA results.
  4. [Section 3.2 and Appendix D.2] The ablation comparing MoLoRAG+ with MoLoRAGFull shows that traversing all pages (instead of the graph-constrained candidate set) improves Recall@3 from 68.87 to 73.64 on MMLongBench and from 68.92 to 72.30 on LongDocURL. The paper attributes this to a 'marginal performance difference' and emphasizes the time saving, but a 4–5 point recall gain is substantial and suggests that the graph construction is dropping relevant pages. The claim that the graph provides a 'high-quality candidate set' is therefore not fully supported. The authors should provide a cost-benefit analysis (e.g., recall versus number of queried pages) that shows the graph-based trade-off is actually favorable, or qualify the claim that the graph preserves accuracy.
  5. [Section 4.3 and Table 3] Several load-bearing design choices are fixed without any sensitivity analysis: the edge threshold theta = 0.4, exploration set size w = 3, hop limit nhop = 4, and the equal weighting in the Combine(·) function. The graph-connectivity assumption is central to the method: if an evidence page is not within a few hops of the initial semantic set, traversal cannot find it. The authors should at least report retrieval performance over a range of theta and w/nhop values on MMLongBench, and justify the chosen weights rather than simply using the average.
minor comments (6)
  1. [Abstract and Section 4.3] The abstract reports average improvements of 9.68% in accuracy and 7.44% in retrieval precision, while Section 4.3 reports 9.94% on MMLongBench and 7.16% on LongDocURL. These numbers should be reconciled or the abstract should cite the specific table.
  2. [Table 1] In Table 1, FetaTab's average tokens are printed as '16,524.5' but the formatting suggests a possible typo ('16.524.5'); please verify.
  3. [Section D.6] The naming of the logical-only variant is inconsistent: 'MoLoRAGLogi' in Table 7, 'MoLoRAGlogi' in the text of D.6. Please unify.
  4. [Appendix D.1] The MRR definition is unclear: the formula as written states MRR@K = 1 if p_i is the first relevant page and 0 otherwise, but the text then says 'where i denotes the position of the first relevant page'. The formula should be MRR@K = 1/i with i being that position, otherwise the metric is mis-specified.
  5. [Figure 1] The example in Figure 1 contains repeated identical thumbnails of '24 Adams Street Combined Sewer System Performance for a Typical Year', which is visually confusing; please clarify which page is the ground-truth evidence page and which is the retrieved page.
  6. [Section 3.2] The notation for the logical relevance score uses both s_logi and s^logi_i; please use a single consistent notation throughout.

Circularity Check

1 steps flagged · score 4.0 of 10

Fine-tuned MoLoRAG+ is trained on snapshots sampled from the same benchmarks used for retrieval evaluation, so its reported gains are not independent predictions.

  1. fitted input called prediction [Appendix C (Training Data) and Section 4.3 (Retrieval Performance Comparison)]
    "Document snapshots are randomly selected from MMLongBench (Ma et al., 2024b) and LongDocURL (Deng et al., 2024), as these datasets contain multi-modal, information-rich documents. ... Since only MMLongBench and LongDocURL provide ground-truth evidence pages for each query, our comparison is confined to these two datasets."

    The fine-tuned retrieval engine is fit on 3,519 page snapshots sampled from the exact two benchmarks on which retrieval and QA performance are then reported. No document-disjoint or page-disjoint split is described anywhere in the paper. Consequently, MoLoRAG+'s higher logical-relevance scores and its top-K retrieval results in Tables 2-3 can be produced by recognizing pages already seen during SFT rather than by logic-aware generalization. The training-free MoLoRAG rows are not affected by this step, but the paper's fine-tuned variant and the abstract's aggregated improvements lose their status as independent predictions.

full rationale

Aside from the fine-tuned variant's train/evaluation overlap, the paper's derivation chain is self-contained. The page graph is built from ColPali embeddings and a fixed similarity threshold, the VLM logical-relevance score is an independent model output, and retrieval accuracy is measured against human-annotated ground-truth evidence pages rather than against the model's own scores. There are no load-bearing self-citations, no imported uniqueness theorems, and no ansatz smuggled in via citation; the graph-connectivity concern raised by the reader is an assumption about the method's coverage, not a circularity. The one substantive circular pattern is MoLoRAG+: its training snapshots are randomly selected from MMLongBench and LongDocURL, the same datasets used for evaluation in Section 4.3, and the paper reports no split that would make the fine-tuned 'prediction' independent. This contaminates the headline fine-tuned numbers while leaving the training-free framework's claims intact, so the overall circularity score is moderate rather than high.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical or mathematical entities. Its parameters are hyperparameters chosen manually, and the main epistemic load is on the VLM's relevance score and the embedding-based page graph.

free parameters (5)
  • Graph edge threshold theta = 0.4
    Edges between pages are added when ColPali embedding similarity exceeds 0.4; set without reported sensitivity analysis (Appendix D.2).
  • Exploration set size w = 3
    Number of pages kept as the exploration set at each traversal step; chosen manually, no tuning analysis (Algorithm 1, Appendix D.2).
  • Hop limit nhop = 4
    Maximum number of graph traversal hops; set in Appendix D.2 without sensitivity study.
  • Combination weights = equal average
    Semantic and logical relevance scores are combined via a simple average; no learned or tuned weighting is reported (Section 3.2, Appendix D.2).
  • Relevance score scale = 1-5 discrete
    The VLM judge rates page relevance on a 1-5 scale defined by the prompt in Appendix B; the granularity is a modeling choice.
assumptions (4)
  • domain assumption ColPali embeddings capture inter-page semantic relationships sufficient to build a useful page graph
    The graph is constructed solely from ColPali embedding similarity (Section 3.2), so the quality of the graph depends on this encoder and the threshold.
  • domain assumption VLM relevance judgments are a reliable proxy for logical relevance
    The retrieval engine scores pages 1-5 via prompt in Appendix B; the paper does not evaluate the VLM's judgment quality against human logical-relevance labels.
  • domain assumption Ground-truth evidence pages in MMLongBench and LongDocURL are complete and correct relevance labels
    Retrieval metrics (Recall, Precision, NDCG, MRR) treat these pages as the only relevant pages (Appendix D.1).
  • domain assumption GPT-4o self-consistency filtering yields valid training labels
    MoLoRAG+ training data is generated by GPT-4o with a self-consistency check (Section 3.3, Appendix C); the paper does not validate these labels against human judgments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MoLoRAG: Bootstrapping Document Understanding via Multi-modal Logic-aware Retrieval." pith.science (2026). https://pith.science/paper/3IYXJQDM

@misc{pith2026250907666,
  author       = {Pith},
  title        = {Pith review of: MoLoRAG: Bootstrapping Document Understanding via Multi-modal Logic-aware Retrieval},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IYXJQDM}},
  note         = {Machine review of arXiv:2509.07666}
}
abstract

Document Understanding is a foundational AI capability with broad applications, and Document Question Answering (DocQA) is a key evaluation task. Traditional methods convert the document into text for processing by Large Language Models (LLMs), but this process strips away critical multi-modal information like figures. While Large Vision-Language Models (LVLMs) address this limitation, their constrained input size makes multi-page document comprehension infeasible. Retrieval-augmented generation (RAG) methods mitigate this by selecting relevant pages, but they rely solely on semantic relevance, ignoring logical connections between pages and the query, which is essential for reasoning. To this end, we propose MoLoRAG, a logic-aware retrieval framework for multi-modal, multi-page document understanding. By constructing a page graph that captures contextual relationships between pages, a lightweight VLM performs graph traversal to retrieve relevant pages, including those with logical connections often overlooked. This approach combines semantic and logical relevance to deliver more accurate retrieval. After retrieval, the top-$K$ pages are fed into arbitrary LVLMs for question answering. To enhance flexibility, MoLoRAG offers two variants: a training-free solution for easy deployment and a fine-tuned version to improve logical relevance checking. Experiments on four DocQA datasets demonstrate average improvements of 9.68% in accuracy over LVLM direct inference and 7.44% in retrieval precision over baselines. Codes and datasets are released at https://github.com/WxxShirley/MoLoRAG.

Figures

Figures reproduced from arXiv: 2509.07666 by the authors.

Figure 1
Figure 1. Illustration of a retrieval example on Long￾DocURL (Deng et al., 2024). Both M3DocRAG (Cho et al., 2024) and MDocAgent (Han et al., 2025) rely solely on semantic relevance between the query and the page for retrieval. As a result, they retrieve a page containing keywords from the question but lacking the necessary information to answer it. In contrast, the ground-truth evidence page, successfully retrieved by our Mo… view at source ↗
Figure 2
Figure 2. Illustration of MoLoRAG framework. a = LVLM(q, Process(P r )). 3.2 Logic-aware Page Retrieval In this subsection, we detail the retrieval process of MoLoRAG. We first construct a page graph as a graph-based index, depicting the relationships between pages within a document. Then, a VLM serves as the retrieval engine, performing reasoning over this graph through traversal to adaptively iden￾tify pages that are both s… view at source ↗
Figure 3
Figure 3. Illustration of training data generation for MoLoRAG+. exploration set for the next iteration. The traversal continues until either the candidate set is empty or the maximum hop limit is reached. Both the explo￾ration set size w and the hop limit nhop constrain the traversal space, ensuring efficiency by avoiding the exhaustive process of sequentially traversing every page in the document. Once the traversal is comp… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Performance trends of MoLoRAG across different top-K retrieval settings for LVLMs. LVLMs with extensive context support (e.g., Qwen2.5-VL series) benefit from retrieving more pages, improving performance with higher K. In contrast, LVLMs with limited context capacity (…
Figure 5
Figure 5. Figure 5: Case study on LongDocURL. MoLoRAG successfully retrieves the correct evidence page for the given question by leveraging logical relevance, enabling it to provide the correct answer. In contrast, both LVLM direct inference and other baseline methods fail to answer the q…
Figure 6
Figure 6. Figure 6: Illustration of retrieval scalability. We present the average number of pages queried by MoLoRAG and the total number of pages across all test samples for each dataset. of queried pages decreases significantly. For in￾stance, fewer than 50% of the pages are queried for…
Figure 8
Figure 8. Figure 8: Case study under the top-3 setting on LongDocURL involving cross-page reasoning. The given question requires evidence from two separate pages to derive the correct answer. LVLM direct inference fails due to its limited context capacity (e.g., processing up to 30 pages)…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 5 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. https://openreview.net/forum?id=hSyW5go0v8 Self- RAG : Learning to retrieve, generate, and critique through self-reflection . In The Twelfth International Conference on Learning Representations

    2024

  4. [4]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, and 8 others. 2025. Qwen2.5-vl technical report. arXiv preprint arXiv:2502.13923

    arXiv 2025

  5. [5]

    Wenhu Chen, Hexiang Hu, Xi Chen, Pat Verga, and William W. Cohen. 2022. https://arxiv.org/abs/2210.02928 Murag: Multimodal retrieval-augmented generator for open question answering over images and text . Preprint, arXiv:2210.02928

    arXiv 2022

  6. [6]

    Zhanpeng Chen, Chengjin Xu, Yiyan Qi, and Jian Guo. 2024. Mllm is a strong reranker: Advancing multimodal retrieval-augmented generation via knowledge-enhanced reranking and noise-injected training. arXiv preprint arXiv:2407.21439

    arXiv 2024

  7. [7]

    Jaemin Cho, Debanjan Mahata, Ozan Irsoy, Yujie He, and Mohit Bansal. 2024. M3docrag: Multi-modal retrieval is what you need for multi-page multi-document understanding. arXiv preprint arXiv:2411.04952

    arXiv 2024

  8. [8]

    DeepSeek-AI. 2025. https://arxiv.org/abs/2412.19437 Deepseek-v3 technical report . Preprint, arXiv:2412.19437

    arXiv 2025

Show all 42 references
  1. [9]

    Chao Deng, Jiale Yuan, Pi Bu, Peijie Wang, Zhong-Zhi Li, Jian Xu, Xiao-Hui Li, Yuan Gao, Jun Song, Bo Zheng, and Cheng-Lin Liu. 2024. https://arxiv.org/abs/2412.18424 Longdocurl: a comprehensive multimodal long document benchmark integrating understanding, reasoning, and locat...

    2024 arXiv

  2. [10]

    Yihao Ding, Zhe Huang, Runlin Wang, YanHang Zhang, Xianru Chen, Yuzhong Ma, Hyunsuk Chung, and Soyeon Caren Han. 2022. V-doc: Visual questions answers with documents. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 21492--21498

    2022

  3. [11]

    Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Dasha Metropolitansky, Robert Osazuwa Ness, and Jonathan Larson. 2025. https://arxiv.org/abs/2404.16130 From local to global: A graph rag approach to query-focused summarization . Prepr...

    2025 arXiv

  4. [12]

    Manuel Faysse, Hugues Sibille, Tony Wu, Bilel Omrani, Gautier Viaud, Céline Hudelot, and Pierre Colombo. 2024. https://arxiv.org/abs/2407.01449 Colpali: Efficient document retrieval with vision language models . Preprint, arXiv:2407.01449

    2024 arXiv

  5. [13]

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, Meng Wang, and Haofen Wang. 2024. https://arxiv.org/abs/2312.10997 Retrieval-augmented generation for large language models: A survey . Preprint, arXiv:2312.10997

    2024 arXiv

  6. [14]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    2024 arXiv

  7. [15]

    Siwei Han, Peng Xia, Ruiyi Zhang, Tong Sun, Yun Li, Hongtu Zhu, and Huaxiu Yao. 2025. Mdocagent: A multi-modal multi-agent framework for document understanding. arXiv preprint arXiv:2503.13964

    2025 arXiv

  8. [16]

    Xiaoxin He, Yijun Tian, Yifei Sun, Nitesh V Chawla, Thomas Laurent, Yann LeCun, Xavier Bresson, and Bryan Hooi. 2024. https://openreview.net/forum?id=MPJ3oXtTZl G-retriever: Retrieval-augmented generation for textual graph understanding and question answering . In The Thirty-e...

    2024

  9. [17]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lo RA : Low-rank adaptation of large language models. In International Conference on Learning Representations

    2022

  10. [18]

    Yulong Hui, Yao Lu, and Huanchen Zhang. 2024. https://openreview.net/forum?id=MS4oxVfBHn UDA : A benchmark suite for retrieval augmented generation in real-world document analysis . In The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track

    2024

  11. [19]

    Albert Qiaochu Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de Las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, L'elio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, T...

    2023 arXiv

  12. [20]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2021. https://arxiv.org/abs/2005.11401 Retrieval-augmented generation for knowledge-int...

    2021 arXiv

  13. [21]

    Bo Li, Kaichen Zhang, Hao Zhang, Dong Guo, Renrui Zhang, Feng Li, Yuanhan Zhang, Ziwei Liu, and Chunyuan Li. 2024 a . https://llava-vl.github.io/blog/2024-05-10-llava-next-stronger-llms/ Llava-next: Stronger llms supercharge multimodal capabilities in the wild

    2024

  14. [22]

    Shilong Li, Yancheng He, Hangyu Guo, Xingyuan Bu, Ge Bai, Jie Liu, Jiaheng Liu, Xingwei Qu, Yangguang Li, Wanli Ouyang, Wenbo Su, and Bo Zheng. 2024 b . https://aclanthology.org/2024.findings-emnlp.746/ G raph R eader: Building graph-based agent to enhance long-context abiliti...

    2024

  15. [23]

    Hao Liu, Zhengren Wang, Xi Chen, Zhiyu Li, Feiyu Xiong, Qinhan Yu, and Wentao Zhang. 2025. https://arxiv.org/abs/2502.12442 Hoprag: Multi-hop reasoning for logic-aware retrieval-augmented generation . Preprint, arXiv:2502.12442

    2025 arXiv

  16. [24]

    Haoyu Lu, Wen Liu, Bo Zhang, Bingxuan Wang, Kai Dong, Bo Liu, Jingxiang Sun, Tongzheng Ren, Zhuoshu Li, Hao Yang, Yaofeng Sun, Chengqi Deng, Hanwei Xu, Zhenda Xie, and Chong Ruan. 2024. https://arxiv.org/abs/2403.05525 Deepseek-vl: Towards real-world vision-language understand...

    2024 arXiv

  17. [25]

    Xueguang Ma, Shengyao Zhuang, Bevan Koopman, Guido Zuccon, Wenhu Chen, and Jimmy Lin. 2024 a . Visa: Retrieval augmented generation with visual source attribution. arXiv preprint arXiv:2412.14457

    2024 arXiv

  18. [26]

    Yubo Ma, Yuhang Zang, Liangyu Chen, Meiqi Chen, Yizhu Jiao, Xinze Li, Xinyuan Lu, Ziyu Liu, Yan Ma, Xiaoyi Dong, Pan Zhang, Liangming Pan, Yu-Gang Jiang, Jiaqi Wang, Yixin Cao, and Aixin Sun. 2024 b . https://arxiv.org/abs/2407.01523 Mmlongbench-doc: Benchmarking long-context ...

    2024 arXiv

  19. [27]

    Ahmed Masry, Xuan Long Do, Jia Qing Tan, Shafiq Joty, and Enamul Hoque. 2022. https://doi.org/10.18653/v1/2022.findings-acl.177 C hart QA : A benchmark for question answering about charts with visual and logical reasoning . In Findings of the Association for Computational Ling...

    2022 doi

  20. [28]

    Minesh Mathew, Dimosthenis Karatzas, and CV Jawahar. 2021. Docvqa: A dataset for vqa on document images. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, pages 2200--2209

    2021

  21. [29]

    Jamshed Memon, Maira Sami, Rizwan Ahmed Khan, and Mueen Uddin. 2020. https://doi.org/10.1109/ACCESS.2020.3012542 Handwritten optical character recognition (ocr): A comprehensive systematic literature review (slr) . IEEE Access, 8:142642--142668

    2020

  22. [30]

    OpenAI. 2024. https://arxiv.org/abs/2410.21276 Gpt-4o system card . Preprint, arXiv:2410.21276

    2024 arXiv

  23. [31]

    Qwen. 2025. https://arxiv.org/abs/2412.15115 Qwen2.5 technical report . Preprint, arXiv:2412.15115

    2025 arXiv

  24. [32]

    Parth Sarthi, Salman Abdullah, Aditi Tuli, Shubh Khanna, Anna Goldie, and Christopher D. Manning. 2024. Raptor: Recursive abstractive processing for tree-organized retrieval. In International Conference on Learning Representations (ICLR)

    2024

  25. [33]

    Jianwei Sun, Chaoyang Mei, Linlin Wei, Kaiyu Zheng, Na Liu, Ming Cui, and Tianyi Li. 2024. https://arxiv.org/abs/2403.09167 Dial-insight: Fine-tuning large language models with high-quality domain-specific data preventing capability collapse . Preprint, arXiv:2403.09167

    2024 arXiv

  26. [34]

    Manan Suri, Puneet Mathur, Franck Dernoncourt, Kanika Goswami, Ryan A Rossi, and Dinesh Manocha. 2024. Visdom: Multi-document qa with visually rich elements using multimodal retrieval-augmented generation. arXiv preprint arXiv:2412.10704

    2024 arXiv

  27. [35]

    Ryota Tanaka, Kyosuke Nishida, Kosuke Nishida, Taku Hasegawa, Itsumi Saito, and Kuniko Saito. 2023. Slidevqa: A dataset for document visual question answering on multiple images. In AAAI

    2023

  28. [36]

    Rubèn Tito, Dimosthenis Karatzas, and Ernest Valveny. 2023. https://arxiv.org/abs/2212.05935 Hierarchical multimodal transformers for multi-page docvqa . Preprint, arXiv:2212.05935

    2023 arXiv

  29. [37]

    Bin Wang, Chao Xu, Xiaomeng Zhao, Linke Ouyang, Fan Wu, Zhiyuan Zhao, Rui Xu, Kaiwen Liu, Yuan Qu, Fukai Shang, Bo Zhang, Liqun Wei, Zhihao Sui, Wei Li, Botian Shi, Yu Qiao, Dahua Lin, and Conghui He. 2024 a . https://arxiv.org/abs/2409.18839 Mineru: An open-source solution fo...

    2024 arXiv

  30. [38]

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. 2024 b . https://arxiv.org/abs/2409.12191 Qwen...

    2024 arXiv

  31. [39]

    Haoran Wei, Chenglong Liu, Jinyue Chen, Jia Wang, Lingyu Kong, Yanming Xu, Zheng Ge, Liang Zhao, Jianjian Sun, Yuang Peng, Chunrui Han, and Xiangyu Zhang. 2024. https://arxiv.org/abs/2409.01704 General ocr theory: Towards ocr-2.0 via a unified end-to-end model . Preprint, arXi...

    2024 arXiv

  32. [40]

    Zhishang Xiang, Chuanjie Wu, Qinggang Zhang, Shengyuan Chen, Zijin Hong, Xiao Huang, and Jinsong Su. 2025. When to use graphs in rag: A comprehensive analysis for graph retrieval-augmented generation. arXiv preprint arXiv:2506.05690

    2025

  33. [41]

    Junyuan Zhang, Qintong Zhang, Bin Wang, Linke Ouyang, Zichen Wen, Ying Li, Ka-Ho Chow, Conghui He, and Wentao Zhang. 2024. Ocr hinders rag: Evaluating the cascading impact of ocr on retrieval-augmented generation. arXiv preprint arXiv:2412.02592

    2024 arXiv

  34. [42]

    Qinggang Zhang, Shengyuan Chen, Yuanchen Bei, Zheng Yuan, Huachi Zhou, Zijin Hong, Junnan Dong, Hao Chen, Yi Chang, and Xiao Huang. 2025. A survey of graph retrieval-augmented generation for customized large language models. arXiv preprint arXiv:2501.13958

    2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.