Pith. sign in

REVIEW 4 major objections 6 minor 48 references

MMESGBench: Pioneering Multimodal Understanding and Complex Reasoning Benchmark for ESG Tasks

T0 review · 4 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 MMESGBench is a new 933-question benchmark for multimodal reasoning over ESG documents, and the strongest system tested answers only about 52 percent of it correctly.

desk verdict MMESGBench is a genuinely useful new resource—the first multimodal QA benchmark for ESG documents—but the leaderboard should be read with caution because the same model family generated, verified, and topped it. read the letter →

arxiv 2507.18932 v2 pith:5MVJWE4T submitted 2025-07-25 cs.MM cs.CL

Lei Zhang , Xin Zhou , Chaoyue He , Di Wang , Yi Wu , Hong Xu , Wei Liu , Chunyan Miao This is my paper · ORCID
classification cs.MMcs.CL
keywords ESGmultimodaldocumentunderstandingbenchmarkdatasetquestionansweringretrieval-augmentedgenerationlayout-awarereasoninglong-documentvision-languagemodels
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

ESG reports matter for sustainability, compliance, and investment, but they are long, multimodal, and structurally varied, and no benchmark was designed for the reasoning they demand. The paper introduces MMESGBench to fill that gap: 933 expert-validated question-answer pairs drawn from 45 real ESG documents covering corporate reports, standards and frameworks, and government and international-organization publications. Each question is labeled as single-page, cross-page, or unanswerable, with fine-grained evidence-type tags covering text, tables, charts, images, and layout. On this testbed, the paper finds that multimodal models and retrieval-augmented pipelines clearly outperform text-only baselines, yet the best configuration, retrieving five pages with ColPali and answering with Qwen-VL-Max, reaches only 51.8 percent accuracy, so the benchmark has substantial room for improvement.

What carries the argument

The load-bearing object is the MMESGBench dataset: 933 QA pairs, each attached to evidence pages, modality tags, and a difficulty score, built so that a model must jointly read text, tables, charts, images, and layout. The generation pipeline uses Qwen-VL-Max to produce candidate pairs from rendered page images, Qwen-Max to check answerability and consistency, and expert reviewers to revise or discard flagged items; cross-page pairs are generated by embedding pages with PaliGemma-3B, grouping them via FAISS-based clustering, and prompting for aggregation, comparison, or causal-linkage questions. Unanswerable items are included deliberately to test whether a model knows when the answer is not in the document.

What would settle it

Re-annotate a random sample of 100 of the 933 QA pairs with two independent ESG-specialist teams, giving them full access to the source documents, and measure agreement on answers, evidence pages, and unanswerable decisions; if agreement is low, or if re-annotation finds a meaningful share of answers or 'unanswerable' labels is wrong, the benchmark cannot support trustworthy model comparisons.

Watch

Extended reading notes

Core claim

The central claim is that no existing benchmark captures the full multimodal and structural complexity of ESG documents, and that MMESGBench is the first dataset to do so, enabling reliable measurement of document-level reasoning. The paper substantiates this with a human-AI collaborative pipeline: a multimodal LLM generates candidate QA pairs from layout-aware page images, an LLM verifier filters weakly grounded or inconsistent items, and ESG experts calibrate the survivors; cross-page questions are produced by embedding pages, clustering them semantically, and prompting for multi-hop synthesis, while unanswerable questions probe hallucination. Initial experiments show that rendering pages as images and adding retrieval substantially improves accuracy over OCR-based text inputs, especially on layout and cross-page questions, and that chart-based reasoning remains the weakest modality.

Load-bearing premise

The load-bearing premise is that the gold answers, including the 'unanswerable' labels, are correct after expert calibration; the paper reports no inter-annotator agreement, no count of expert revisions, and relies on an LLM to flag errors, so if many answers are wrong or some unanswerable items do have evidence, the benchmark's model rankings could be artifacts.

Editorial extensions

If this is right

  • Multimodal models and retrieval-augmented pipelines become the necessary baseline for ESG document QA, while text-only OCR baselines fall far behind and should be treated as the floor.
  • The benchmark's modality and location splits let researchers pinpoint failure modes: charts and layout-sensitive questions are the hard cases, while pure text and layout cues are comparatively easy.
  • Unanswerable questions turn hallucination resistance into a measurable quantity, and models that simply guess 'not answerable' inflate accuracy but drop in F1, so rankings must consider both metrics.
  • With the strongest system at about 52 percent accuracy, the benchmark offers a concrete yardstick for progress in long-context, layout-aware, multimodal document reasoning.

Reading between the lines

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

  • A natural follow-up the paper does not run is to have two independent expert teams re-label a random subset of the 933 pairs and measure inter-annotator agreement; that would tell how much of the benchmark's validity rests on the single LLM-flag-to-expert review chain.
  • Because the construction pipeline is domain-agnostic, the same generate-verify-calibrate recipe could be transplanted to other long regulated documents, such as financial filings or clinical trial reports, where layout and tables also carry meaning.
  • The large gap between open-source multimodal models and the proprietary Qwen-VL-Max points to a concrete development target: improving open models' numeric and spatial reasoning on charts, which the paper identifies as the weakest area.
  • The unanswerable-question design could be reused as a cheap hallucination probe for arbitrary long-document QA systems, not just ESG ones.
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

4 major / 6 minor

Summary. The paper introduces MMESGBench, a multimodal question-answering benchmark for ESG documents, built from 45 ESG reports and standards and containing 933 QA pairs labeled as single-page, cross-page, or unanswerable. The construction pipeline uses Qwen-VL-Max to generate candidate QA pairs from rendered document pages, Qwen-Max to perform automated quality control and flag mismatches, and ESG experts to review flagged items. The authors evaluate 15 text-only, multimodal, and retrieval-augmented models, reporting that multimodal and RAG pipelines substantially outperform text-only baselines, with ColPali+Qwen-VL-Max reaching 51.8% accuracy. The dataset and code are released publicly.

Significance. If the gold labels and rankings are trustworthy, MMESGBench would be a useful first benchmark for multimodal, long-document ESG reasoning, with clear practical applications in compliance, disclosure validation, and RAG evaluation. Strengths include the diversity of document sources and types, the fine-grained evidence and modality annotations, the inclusion of unanswerable questions, and the public release of the dataset. However, the current evaluation loop is not independent of the data-generation process: the same model family that generates and verifies the QA pairs also achieves the top scores. The paper also lacks the inter-annotator agreement, filtering statistics, and significance testing needed to support the reliability claims. The benchmark's value depends on establishing label independence and fairer comparisons, so the central claims are not yet established.

major comments (4)
  1. [§3.3 and Table 1] The data-creation and evaluation loop is not independent. Qwen-VL-Max generates the candidate QA pairs, Qwen-Max performs the automated quality control, and only flagged items are reviewed by experts; then Qwen-VL-Max achieves the best non-RAG accuracy (40.0) and ColPali+Qwen-VL-Max the best overall (51.8). Because the verifier belongs to the same model family, systematic errors in reading tables or charts or in answer style will tend to agree with the generator and therefore escape flagging, so the gold labels and the leaderboard partly measure self-consistency with Qwen's answer distribution rather than general ESG understanding. The paper reports no inter-annotator agreement, no count of expert revisions or discards, and no error-rate estimate on non-flagged items. Please report these statistics, audit a random sample of non-flagged QA pairs with independent annotators, use multiple diverse verifier models, and report results for models outside the generating family separately.
  2. [§3.2, unanswerable QA] The construction of unanswerable questions is not adequately validated. The paper states that these items are generated 'while ensuring no corresponding evidence exists,' but this guarantee comes only from an LLM prompt; no document-level retrieval or expert verification is described to confirm that evidence is absent from the full document, which can exceed 2,000 pages. Since unanswerable items constitute 15.9% of the benchmark and Table 1 shows large differences across models on the UNA column, any mislabeled negatives would directly distort accuracy and F1. Please describe and implement a validation protocol for unanswerable items using exhaustive or retrieval-assisted evidence search with human confirmation, and report the outcome.
  3. [§4.1 and Table 1] The headline comparison between text-only and multimodal or RAG pipelines is confounded by differences in input construction. Text-only LLMs receive truncated OCR up to 120 pages with no retrieval, while multimodal models receive page images and RAG pipelines receive retrieved pages; this varies the context budget, access to retrieval, and the decoder simultaneously. Therefore the claims that 'multimodal models substantially outperform their text-only counterparts' and that 'ColPali+Qwen-VL-Max improves overall accuracy by 30%' are not cleanly attributable to multimodality or retrieval. Please evaluate matched settings, such as OCR with ColBERT retrieval for text decoders, the same decoder in text-only and multimodal modes, and equal context lengths, and report confidence intervals or significance tests for the accuracy differences.
  4. [§3.3, quality-control pipeline] The paper gives no quantitative account of the quality-control pipeline: how many candidate QA pairs were generated, how many were removed by no-context inference, how many were flagged by the automated verifier, how many were discarded or revised by experts, and how many experts or annotators participated. Without these numbers, the claim that the dataset is 'high-fidelity' and 'evaluation-ready' cannot be assessed. Please report these pipeline statistics and, if possible, expert agreement coefficients on a sample of the final QA pairs.
minor comments (6)
  1. [§1] There is a typo in 'the fellowing three interrelated dimensions'; it should be 'following'.
  2. [Table 1] In the Mistral-Instruct-v0.1 row, the values appear as '10.684.522.2' instead of separate entries; please fix the formatting so all numbers are readable as distinct metrics.
  3. [Figure 2] Figure 2(a) is referenced as showing dataset statistics, but the panel content is not displayed or described in the text; please ensure the panel and its contents are included and explained.
  4. [§4.1] The paper mentions 'Qwen-VL-7B' in the evaluation but cites Qwen2-VL; please clarify the exact model version and configuration for reproducibility.
  5. [§4.1] The automatic short-form answer extraction and rule-based score computation are described only briefly; please provide details or pseudocode for how partial matches are scored and how abstentions are handled in the F1 metric.
  6. [§1] The phrase 'stakeholder-orientated' should be 'stakeholder-oriented'; similar minor language issues appear throughout and should be corrected.

Circularity Check

2 steps flagged · score 5.0 of 10

MMESGBench's gold labels are generated and verified by Qwen-family models, and Qwen-VL-Max is then reported as the top performer; the headline result therefore partly measures self-consistency, with expert review only on flagged items and no quantified human agreement.

  1. self definitional [Sections 3.2, 3.3, and 4.2; Table 1]
    "For single-page QA generation, each document page is rendered as an image and processed by Qwen-VL-max [40] ... For each remaining QA pair, we re-infer the answer by providing the evidence page and question to multimodal LLM. The model's predicted answer is then compared with the annotated reference, and samples with low token-level F1 score or semantic mismatch are flagged ... For instance, Qwen-VL-Max achieves significantly stronger results than Qwen-Max, yielding over 60% relative improvement in accuracy"

    The benchmark's reference answers are definitionally produced by Qwen-VL-Max, and the automated verifier that decides which items are flagged for expert review is the same Qwen family (Qwen-Max, then a multimodal LLM). Non-flagged items are never independently checked, so systematic Qwen-family answer styles and misreadings can enter the gold set unchecked. Table 1 then reports Qwen-VL-Max, and ColPali+Qwen-VL-Max, as the best performers. A model sharing the generator's biases will agree with the gold labels more often, so the claim that multimodal and RAG models substantially outperform text-only baselines is partly a self-consistency effect rather than purely task mastery.

  2. self definitional [Section 3.2 and Table 1 (unanswerable QA generation)]
    "To assess model robustness, we also include unanswerable QA pairs, generated at the document- or section-level by prompting the model to produce plausible yet unsupported questions aligned with the document's theme, while ensuring no corresponding evidence exists."

    The unanswerable labels are created by prompting Qwen-VL-Max to decide what is unsupported; the quality-control step re-infers answers for remaining QA pairs, but no independent exhaustive evidence search is reported for these negative examples. The UNA column in Table 1 therefore partly measures how well a model reproduces Qwen's own notion of answerability, rather than general hallucination resistance. For models from the same family, high UNA accuracy is expected by construction; the paper does not provide external verification that every 'unanswerable' item truly lacks evidence anywhere in the document.

full rationale

The central contribution is a dataset and an evaluation protocol, not a derived physical law, so the usual fit-versus-prediction circularity does not apply verbatim. However, the headline experimental claim—that multimodal and retrieval-augmented models substantially outperform text-only baselines, with Qwen-VL-Max as the strongest model—is not independent of the data-creation process. Section 3.2 makes Qwen-VL-Max the author of the gold answers; Section 3.3 uses Qwen-Max as the automated gatekeeper, with expert review applied only to flagged items and no reported inter-annotator agreement or revision statistics. Under those conditions, a high Qwen-VL-Max score is partly self-consistency: the reference answers and the evaluated model share the same systematic reading of charts, tables, and layouts, so shared errors will not be flagged. The paper is not wholly circular—evaluation still requires the model to retrieve evidence and produce exact-match/F1 answers, and experts did review flagged examples—but the missing quantification prevents separating genuine multimodal competence from generator/verifier agreement. No load-bearing self-citations or imported uniqueness theorems were found; reference [16] is merely contextual. Overall, this is a moderate, partial circularity rather than a forced derivation.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters are fitted to data for the central dataset claim. The main unstated premises are about gold-label correctness, the validity of LLM-generated unanswerable questions, and the fairness of the text-only evaluation protocol.

assumptions (3)
  • domain assumption LLM-generated QA pairs, after automated filtering and expert review, constitute correct ground truth for evaluating models.
    The benchmark's validity rests on gold answers being accurate; the paper does not report inter-annotator agreement or the number and type of expert corrections.
  • ad hoc to paper Unanswerable questions generated by prompting an LLM are guaranteed to have no evidence in the document.
    Section 3.2: 'unanswerable QA pairs ... generated ... by prompting the model to produce plausible yet unsupported questions, while ensuring no corresponding evidence exists.' No independent verification is described beyond LLM checks.
  • domain assumption Document pages rendered as images preserve all semantic content needed for QA, and OCR text extraction is a fair representation for text-only models.
    Evaluation protocol in Section 4.1: text-only models use OCR-extracted text truncated to context limits; any OCR errors or loss of layout information may disadvantage them.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MMESGBench: Pioneering Multimodal Understanding and Complex Reasoning Benchmark for ESG Tasks." pith.science (2026). https://pith.science/paper/5MVJWE4T

@misc{pith2026250718932,
  author       = {Pith},
  title        = {Pith review of: MMESGBench: Pioneering Multimodal Understanding and Complex Reasoning Benchmark for ESG Tasks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5MVJWE4T}},
  note         = {Machine review of arXiv:2507.18932}
}
read the original abstract

Environmental, Social, and Governance (ESG) reports are essential for evaluating sustainability practices, ensuring regulatory compliance, and promoting financial transparency. However, these documents are often lengthy, structurally diverse, and multimodal, comprising dense text, structured tables, complex figures, and layout-dependent semantics. Existing AI systems often struggle to perform reliable document-level reasoning in such settings, and no dedicated benchmark currently exists in ESG domain. To fill the gap, we introduce \textbf{MMESGBench}, a first-of-its-kind benchmark dataset targeted to evaluate multimodal understanding and complex reasoning across structurally diverse and multi-source ESG documents. This dataset is constructed via a human-AI collaborative, multi-stage pipeline. First, a multimodal LLM generates candidate question-answer (QA) pairs by jointly interpreting rich textual, tabular, and visual information from layout-aware document pages. Second, an LLM verifies the semantic accuracy, completeness, and reasoning complexity of each QA pair. This automated process is followed by an expert-in-the-loop validation, where domain specialists validate and calibrate QA pairs to ensure quality, relevance, and diversity. MMESGBench comprises 933 validated QA pairs derived from 45 ESG documents, spanning across seven distinct document types and three major ESG source categories. Questions are categorized as single-page, cross-page, or unanswerable, with each accompanied by fine-grained multimodal evidence. Initial experiments validate that multimodal and retrieval-augmented models substantially outperform text-only baselines, particularly on visually grounded and cross-page tasks. MMESGBench is publicly available as an open-source dataset at https://github.com/Zhanglei1103/MMESGBench.

Figures

Figures reproduced from arXiv: 2507.18932 by the authors.

Figure 1
Figure 1. The human-AI collaborative multi-stage QA generation framework. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of MMESGBench. (a) Dataset statistics. (b) Distribution of QA instances and total pages per document type; (c) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Comparative analysis of multimodal QA model [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 22 canonical work pages

  1. [1]

    Shafiuddin Rehan Ahmed, Ankit Parag Shah, Quan Hung Tran, Vivek Khetan, Sukryool Kang, Ankit Mehta, Yujia Bao, and Wei Wei. 2025. Enhancing Retrieval for ESGLLM via ESG-CID–A Disclosure Content Index Finetuning Dataset for Mapping GRI and ESRS.arXiv preprint arXiv:2503.10674(2025)

    work page Pith review arXiv 2025

  2. [2]

    Muhammad Arslan, Saba Munawar, and Zainab Riaz. 2024. Sustainable Urban Water Decisions using Generative Artificial Intelligence. In2024 International Conference on Decision Aid Sciences and Applications (DASA). IEEE, 1–5

    work page 2024

  3. [3]

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, et al. 2023. Qwen technical report.arXiv preprint arXiv:2309.16609(2023)

    arXiv 2023

  4. [4]

    Lucas Beyer, Andreas Steiner, André Susano Pinto, Alexander Kolesnikov, Xiao Wang, Daniel Salz, Maxim Neumann, Ibrahim Alabdulmohsin, Michael Tschan- nen, Emanuele Bugliarello, et al. 2024. Paligemma: A versatile 3b vlm for transfer. arXiv preprint arXiv:2407.07726(2024)

    arXiv 2024

  5. [5]

    Xiao Bi, Deli Chen, Guanting Chen, Shanhuang Chen, Damai Dai, Chengqi Deng, Honghui Ding, Kai Dong, Qiushi Du, Zhe Fu, et al. 2024. Deepseek llm: Scaling open-source language models with longtermism.arXiv preprint arXiv:2401.02954 (2024)

    arXiv 2024

  6. [6]

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, et al . 2024. Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 24185–24198

    2024

  7. [7]

    Chao Deng, Jiale Yuan, Pi Bu, Peijie Wang, Zhong-Zhi Li, Jian Xu, Xiao-Hui Li, Yuan Gao, Jun Song, Bo Zheng, et al. 2024. LongDocURL: a Comprehensive Multimodal Long Document Benchmark Integrating Understanding, Reasoning, and Locating.arXiv preprint arXiv:2412.18424(2024)

    arXiv 2024

  8. [8]

    Xiaoyi Dong, Pan Zhang, Yuhang Zang, Yuhang Cao, Bin Wang, Linke Ouyang, Xilin Wei, Songyang Zhang, Haodong Duan, Maosong Cao, et al. 2024. Internlm- xcomposer2: Mastering free-form text-image composition and comprehension in vision-language large model.arXiv preprint arXiv:2401.16420(2024)

    arXiv 2024

Show all 48 references
  1. [9]

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre-Emmanuel Mazaré, Maria Lomeli, Lucas Hosseini, and Hervé Jégou. 2024. The faiss library.arXiv preprint arXiv:2401.08281(2024)

    2024 arXiv

  2. [10]

    Manuel Faysse, Hugues Sibille, Tony Wu, Bilel Omrani, Gautier Viaud, Céline Hudelot, and Pierre Colombo. [n. d.]. Colpali: Efficient document retrieval with vision language models. InThe Thirteenth International Conference on Learning Representations

  3. [11]

    Financial Stability Board. 2017. Task Force on Climate-related Financial Disclo- sures Recommendations. https://www.fsb-tcfd.org/recommendations/

    2017

  4. [12]

    Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. 2024. Chatglm: A fam- ily of large language models from glm-130b to glm-4 all tools.arXiv preprint arXiv:2406.12793(2024)

    2024 arXiv

  5. [13]

    Global Reporting Initiative. 2023. GRI Standards. https://www.globalreporting. org/how-to-use-the-gri-standards/gri-standards-english-language/

    2023

  6. [14]

    Greenhouse Gas Protocol. 2010. GHG. https://ghgprotocol.org/standards- guidance

    2010

  7. [15]

    Tanay Kumar Gupta, Tushar Goel, Ishan Verma, Lipika Dey, and Sachit Bhardwaj

  8. [16]

    Chaoyue He, Xin Zhou, Yi Wu, Xinjia Yu, Yan Zhang, Lei Zhang, Di Wang, Shengfei Lyu, Hong Xu, Xiaoqiao Wang, et al . 2025. ESGenius: Benchmark- ing LLMs on Environmental, Social, and Governance (ESG) and Sustainability Knowledge.arXiv preprint arXiv:2506.01646(2025)

    2025

  9. [17]

    International Organization for Standardization. 2016. ISO Standards. https: //www.iso.org/standards.html

    2016

  10. [18]

    IPCC. 2023. Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. https://www.ipcc.ch/assessment-report/ar6/

    2023

  11. [19]

    Pranab Islam, Anand Kannappan, Douwe Kiela, Rebecca Qian, Nino Scherrer, and Bertie Vidgen. 2023. Financebench: A new benchmark for financial question answering.arXiv preprint arXiv:2311.11944(2023)

    2023 arXiv

  12. [20]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, De- vendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thoma...

    2023 arXiv

  13. [21]

    Omar Khattab and Matei Zaharia. 2020. Colbert: Efficient and effective passage search via contextualized late interaction over bert. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 39–48

    2020

  14. [22]

    Oskar Krabbe, Giel Linthorst, Kornelis Blok, Wina Crijns-Graus, Detlef P Van Vu- uren, Niklas Höhne, Pedro Faria, Nate Aden, and Alberto Carrillo Pineda. 2015. Aligning corporate greenhouse-gas emissions targets with climate goals.Nature Climate Change5, 12 (2015), 1057–1060

    2015

  15. [23]

    Haein Lee, Jang Hyun Kim, and Hae Sun Jung. 2025. ESG-KIBERT: A new paradigm in ESG evaluation using NLP and industry-specific customization. Decis. Support Syst.193 (2025), 114440

    2025

  16. [24]

    Haoyu Lu, Wen Liu, Bo Zhang, Bingxuan Wang, Kai Dong, Bo Liu, Jingxiang Sun, Tongzheng Ren, Zhuoshu Li, Hao Yang, et al. 2024. Deepseek-vl: towards real- world vision-language understanding.arXiv preprint arXiv:2403.05525(2024)

    2024 arXiv

  17. [25]

    Yubo Ma, Yuhang Zang, Liangyu Chen, Meiqi Chen, Yizhu Jiao, Xinze Li, Xinyuan Lu, Ziyu Liu, Yan Ma, Xiaoyi Dong, et al. 2025. MMLONGBENCH-DOC: Bench- marking Long-context Document Understanding with Visualizations.Advances in Neural Information Processing Systems37 (2025), 95963–96010

    2025

  18. [26]

    Ahmed Masry, Xuan Long Do, Jia Qing Tan, Shafiq Joty, and Enamul Hoque. 2022. ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning. InFindings of the Association for Computational Linguistics: ACL 2022. 2263–2279

    2022

  19. [27]

    Minesh Mathew, Viraj Bagal, Rubèn Tito, Dimosthenis Karatzas, Ernest Valveny, and CV Jawahar. 2022. Infographicvqa. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 1697–1706

    2022

  20. [28]

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

    2021

  21. [29]

    Nitesh Methani, Pritha Ganguly, Mitesh M Khapra, and Pratyush Kumar. 2020. Plotqa: Reasoning over scientific plots. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. 1527–1536

    2020

  22. [30]

    Organisation for Economic Co-operation and Development (OECD). 2024. OECD Guidelines. https://www.oecd.org/en.html

    2024

  23. [31]

    Social Accountability International. 2016. SA8000®Standard. https://sa-intl.org/ programs/sa8000/

    2016

  24. [32]

    Ningyuan Sun, Xuefeng Yang, and Yunfeng Liu. 2020. Tableqa: a large-scale chinese text-to-sql dataset for table-aware sql generation.arXiv preprint arXiv:2006.06434(2020)

    2020 arXiv

  25. [33]

    Sustainability Accounting Standards Board. 2023. SASB Standards. https://sasb. org/standards/

    2023

  26. [34]

    Ryota Tanaka, Kyosuke Nishida, Kosuke Nishida, Taku Hasegawa, Itsumi Saito, and Kuniko Saito. 2023. Slidevqa: A dataset for document visual question an- swering on multiple images. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 37. 13636–13645

    2023

  27. [35]

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupati- raju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, et al. 2024. Gemma: Open models based on gemini research and technology. arXiv preprint arXiv:2403.08295(2024)

    2024 arXiv

  28. [36]

    David Thulke, Yingbo Gao, Petrus Pelser, Rein Brune, Rricha Jalota, Floris Fok, Michael Ramos, Ian van Wyk, Abdallah Nasir, Hayden Goldstein, et al . 2024. Climategpt: Towards ai synthesizing interdisciplinary research on climate change. arXiv preprint arXiv:2401.09646(2024)

    2024 arXiv

  29. [37]

    Rubèn Tito, Dimosthenis Karatzas, and Ernest Valveny. 2023. Hierarchical multi- modal transformers for multipage docvqa.Pattern Recognition144 (2023), 109834

    2023

  30. [38]

    United Nations. 2015. Transforming Our World: The 2030 Agenda for Sustainable Development. https://sdgs.un.org/2030agenda

    2015

  31. [39]

    Jordy Van Landeghem, Rubèn Tito, Łukasz Borchmann, Michał Pietruszka, Pawel Joziak, Rafal Powalski, Dawid Jurkiewicz, Mickaël Coustaty, Bertrand Anckaert, Ernest Valveny, et al. 2023. Document understanding dataset and evaluation (dude). InProceedings of the IEEE/CVF Internati...

    2023

  32. [40]

    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. Qwen2-VL: Enhancing Vision-Language Mode...

    2024 arXiv

  33. [41]

    Qilong Wu, Xiaoneng Xiang, Hejia Huang, Xuan Wang, Yeo Wei Jie, Ranjan Sata- pathy, Bharadwaj Veeravalli, et al. 2024. SusGen-GPT: A Data-Centric LLM for Fi- nancial NLP and Sustainability Report Generation.arXiv preprint arXiv:2412.10906 (2024)

    2024 arXiv

  34. [42]

    Yuping Xiao and Li Xiao. 2025. The impact of artificial intelligence-driven ESG performance on sustainable development of central state-owned enterprises listed companies.Scientific Reports15, 1 (2025), 8548. MM ’25, October 27–31, 2025, Dublin, Ireland Lei Zhang et al

    2025

  35. [43]

    Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Haoyu Li, Weilin Zhao, Zhihui He, Qianyu Chen, Huarong Zhou, Zhensheng Zou, Haoye Zhang, Shengding Hu, Zhi Zheng, Jie Zhou, Jie Cai, Xu Han, Guoyang Zeng, Dahai Li, Zhiyuan Liu, and Maosong Sun. 2...

    2024 arXiv

  36. [44]

    Mengdi Zhang, Qiao Shen, Zhiheng Zhao, Shuaian Wang, and George Q Huang

  37. [45]

    Xin Zhou, Lei Zhang, Honglei Zhang, Yixin Zhang, Xiaoxiong Zhang, Jie Zhang, and Zhiqi Shen. 2024. Advancing sustainability via recommender systems: a survey.arXiv preprint arXiv:2411.07658(2024)

    2024 arXiv

  38. [46]

    Yi Zou, Mengying Shi, Zhongjie Chen, Zhu Deng, ZongXiong Lei, Zihan Zeng, Shiming Yang, Hongxiang Tong, Lei Xiao, and Wenwen Zhou. 2025. ESGReveal: An LLM-based approach for extracting structured data from ESG reports.Journal of Cleaner Production489 (2025), 144572

    2025

  39. [2024]

    Knowledge Graph aided LLM based ESG Question-Answering from News. (2024)

    2024

  40. [2025]

    Optimizing ESG reporting: Innovating with E-BERT models in nature language processing.Expert systems with applications265 (2025), 125931

    2025

Pith tools

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