Pith. sign in

REVIEW 5 major objections 4 minor 49 references

SAIL: Sample-Centric In-Context Learning for Document Information Extraction

T0 review · 5 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read No-training document extraction nearly matches fine-tuned models

desk verdict SAIL is a solid, well-ablated extension of ICL-D3IE, but the unreported OCR setup and test-set-tuned hyperparameters mean the headline F1 numbers shouldn't be taken at face value. read the letter →

arxiv 2412.17092 v1 pith:OVS3DCC6 submitted 2024-12-22 cs.CL cs.AI

Jinyu Zhang , Zhiyuan You , Jize Wang , Xinyi Le This is my paper · ORCID
classification cs.CLcs.AI
keywords documentinformationextractionin-contextlearninglargelanguagemodelsvisuallyrichdocumentslayoutsimilarityentity-leveltextsample-centricpromptingtraining-free
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 tries to establish that a training-free, in-context learning approach can extract structured information from document images almost as accurately as fully supervised models. It claims that the key is to build a custom prompt for each test document, pulling in examples that resemble it in three ways: same document-level text, same entity-level text, and same spatial layout. On three benchmarks the method reaches F1 scores of 96.41 on CORD and 98.18 on SROIE with GPT-4, beating the previous in-context baseline and approaching fully trained LayoutLMv3. If correct, this means strong document extraction is possible without task-specific training, using only a handful of hand-picked examples.

What carries the argument

The load-bearing mechanism is a three-way similarity retrieval feeding a single prompt template. Document-level text similarity uses Sentence-BERT embeddings of concatenated OCR text; entity-level similarity embeds individual text blocks (excluding pure numbers) and retrieves nearest entities; layout similarity renders the OCR bounding boxes as a binary image, crops to the content area, resizes, and compares with mean squared error, taking the inverse as the similarity score. These three selections populate a template with candidate label descriptions, entity demonstrations, a layout-analysis step, and document demonstrations. The boxes supplied to the LLM are cropped to the content region, and the layout analysis step asks the LLM to state where each label sits, which the paper claims helps the model transfer layout knowledge to the test document.

What would settle it

One concrete check is to run SAIL on a benchmark while jittering the OCR bounding boxes (for example, shifting each box by a few pixels or removing boxes below a size threshold). If F1 stays essentially unchanged, layout similarity is not doing the claimed work; if it drops sharply, the method is hostage to OCR box quality. Alternatively, replace the layout-similar examples with random documents while keeping the same prompt template; if F1 does not fall, the layout retrieval is superfluous.

Watch

Extended reading notes

Core claim

The central claim is that sample-centric selection of in-context examples unlocks large language models for document information extraction. The authors argue that previous ICL methods fail because they use fixed examples picked by document-level text similarity alone; SAIL instead retrieves, for every test document, layout-similar documents, entity-level text-similar entities, and document-level text-similar documents, then packs them into a unified prompt template. With this recipe, GPT-4 (specifically GPT-4o) reaches 96.41 F1 on CORD and 98.18 on SROIE, close to the fully trained LayoutLMv3 (96.56 and 96.89 on the same sets), and the method outperforms the ICL-D3IE baseline across GPT-3.5, GPT-4, and ChatGLM3. The paper's claim is that the three-way retrieval plus the unified template is what produces this result, not the particular LLM.

Load-bearing premise

The load-bearing premise is that the OCR-derived text and bounding boxes are accurate and that comparing binary images of those boxes by mean squared error captures the layout information that actually determines the entity labels; the paper never names its OCR system.

Editorial extensions

If this is right

  • Document information extraction no longer requires task-specific fine-tuning; few-shot prompting with retrieved examples can rival fully trained extractors.
  • Because the method works across GPT-3.5, GPT-4, and ChatGLM3, the main lever for DIE performance may be in-context prompt construction rather than model scale.
  • The unified template transfers across datasets and label sets with only the candidate-label description changed, suggesting a single prompt recipe for many extraction tasks.
  • Ablations show each similarity type helps on different datasets (entity-level for long forms, layout for receipts), so future ICL systems should mix retrieval signals rather than rely on one.

Reading between the lines

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

  • The layout-similarity step depends on clean OCR boxes; if a deployed OCR produces noisy or incomplete boxes, the binary layout images may mislead retrieval, so the method's gains may shrink outside benchmark OCR conditions.
  • The approach could extend to table extraction or form understanding in other languages, since the prompt template and retrieval are language-agnostic as long as the embedding model supports the language.
  • A testable extension is to combine SAIL's retrieval with multimodal LLMs that take images directly, which might close the remaining gap to full training without OCR at all.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The paper proposes SAIL, a training-free in-context learning method for document information extraction (DIE). SAIL constructs a sample-specific prompt for each test document by selecting three types of demonstrations: document-level text-similar examples, entity-level text-similar examples, and layout-similar examples where layout similarity is computed as the inverse MSE between binary layout images rendered from OCR bounding boxes. The prompt template combines these demonstrations with candidate labels, layout analysis, and the test question in OCR text-plus-box form. Experiments on FUNSD, CORD, and SROIE with ChatGLM3, GPT-3.5, and GPT-4 (gpt-4o) report consistent F1 gains over the ICL-D3IE baseline and, for GPT-4, F1 scores close to full-training methods. The authors include ablations of each component, example order, layout-similarity metric, resize method, and entity representation, plus a Wilcoxon significance test against ICL-D3IE.

Significance. If the reported results hold, SAIL would be a useful contribution: it is a simple, training-free, sample-adaptive ICL method with public code, and it shows consistent gains over the established ICL-D3IE baseline across three benchmarks and three backbone models, together with a thorough ablation study. The Wilcoxon signed-rank test (Appendix C) is a welcome addition, and the synthetic-data experiments (Appendix B) attempt to decompose text versus layout contributions. However, the validity of the empirical claims currently rests on several load-bearing evaluation choices: the OCR inputs are not specified, hyperparameters are selected using test sets, the Table 2 precision/recall values for SAIL are implausible, and the multimodal comparison is not input-controlled. These issues need to be resolved before the central claim can be taken at face value.

major comments (5)
  1. [§3.1, Eq. (1) and §4.1] The OCR system that produces T and B is never identified. The manuscript states only that T and B are "recognized from I by an OCR system" (Eq. 1), and all subsequent similarity computations (Eqs. 2–4) and prompt construction (Eqs. 5–9) depend on these inputs. The comparisons in Table 1 against ICL-D3IE and full-training methods are only meaningful if all methods consume the same T and B. If SAIL uses ground-truth or higher-quality OCR while baselines use noisier OCR, the reported F1 gaps could be input artifacts rather than effects of example selection. Please specify the OCR engine for each dataset, state whether T and B come from ground-truth annotations, and re-run or re-derive baseline comparisons under identical inputs, or otherwise quantify sensitivity of the results to OCR quality.
  2. [Table 2, §4.3] For CORD and FUNSD, the SAIL row reports precision = recall = F1 (96.41/96.41/96.41 and 84.67/84.67/84.67). This exact equality is implausible for entity-level evaluation and suggests a reporting error. Since Table 2 is the basis for the claim that SAIL significantly surpasses multimodal LLMs, the precision/recall values need to be corrected or, if they are genuinely identical, justified with the evaluation formula used.
  3. [Appendix B, Tables A3–A7 and Figure A1] Key design choices are selected using the test sets: the layout similarity metric (Table A3), resize method (Table A4), number of document-level examples (Figure A1), representation of entity examples (Table A6), and number of entity examples (Table A7) are all chosen by comparing F1 on CORD and/or FUNSD test sets. The main paper does not mention a validation split. This makes the reported numbers test-set-tuned, so the headline F1 scores (e.g., 95.80 on CORD with GPT-3.5) may overstate generalization. Please move hyperparameter selection to a validation split or report the selection procedure explicitly and, if feasible, the corresponding validation performance.
  4. [§4.3, Table 2] The multimodal comparison is not input-controlled: SAIL is given OCR-derived text and boxes, whereas GPT-4o and LLaVA-1.5 receive only document images. The comparison conflates the selection/prompt design with the input representation. A controlled comparison (e.g., giving GPT-4o the same OCR text and boxes, or giving SAIL the image) is needed to claim that the method, rather than the input format, is responsible for the gap. At minimum, the paper should discuss this confound explicitly.
  5. [§4.4 and Appendix B] The paper reports single runs without error bars, and it invokes "inherent randomness of LLM generation" to explain a counterintuitive ablation result (Section 4.4, FUNSD #0 vs. #1). This acknowledges run-to-run variability. To support the claim that small differences (e.g., Table 3 adaptive examples, Table 4 example order) are meaningful, the authors should report means and standard deviations over multiple runs, or at least fix and report seeds for the local models, and provide confidence intervals for the main comparisons.
minor comments (4)
  1. [§4.1] The sentence "In the case of GPT-4o, we only provide text prompts as input, while also testing its multimodal capabilities by providing document images and clear task instructions" is ambiguous: it is unclear whether the GPT-4o rows in Table 2 are the text-only or image-based condition. Please clarify which setting produced the reported numbers.
  2. [Figure A3] The caption contains a typo: "Grean" should be "Green".
  3. [Appendix B] The phrase "pulling into a one-dimensional vector" (Effect of the Layout Similarity Comparison Method) would be clearer as "flattening into a one-dimensional vector".
  4. [§4.1 and Table 1] The paper alternates between "GPT-4", "gpt-4o", and "GPT-4 (gpt-4o API version)". Please standardize the naming so readers know which model is being reported in each table.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SAIL's reported F1 scores are genuine LLM inference outputs; the similarity scores in Eqs. (2)-(4) only select training demonstrations and do not determine the predicted labels by construction.

full rationale

The derivation chain is self-contained with respect to circularity. The core claim is that retrieval of layout-similar, entity-similar, and document-similar training examples, assembled into a unified prompt, improves LLM extraction F1. The selection scores in Eqs. (2)-(4) are defined on inputs (Sentence-BERT embeddings of entity/document texts and MSE over rendered box images) and are used solely to choose training examples. The predicted labels Ypred are produced by the LLM in Eq. (9): P(Y|T,B) = (1/ne) Σ PLM(lk | Ccl, Cet, Cl, Cdt, φ(T,B)). No equation defines the target labels as a function of the similarity scores; the training labels enter only as in-context demonstrations, which is standard ICL rather than a fitted parameter renamed as a prediction. The F1 numbers in Tables 1, 2, and A1-A7 are measured against ground-truth test labels, not recovered from the retrieval mechanism. The paper does contain a methodological weakness: several design choices (layout metric, resize method, number of examples, representation format, box source) are selected by comparing F1 on the same CORD and FUNSD test sets, and the OCR system producing T and B is never named, so the comparison against ICL-D3IE may not be input-controlled. These are experimental-validity concerns, not circularity: they do not make any claimed derivation equivalent to its inputs by definition. The only self-citations (e.g., Wang et al. 2023b) appear in related-work enumerations and are not load-bearing. No uniqueness theorem, ansatz-via-citation, or renamed-known-result pattern is present. Accordingly, no circular step is exhibited and the score is 0.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The free parameters are all tuned on the benchmark test sets, which is a mild form of fitting. The method introduces no new entities; its effect relies on standard embedding and image-comparison assumptions.

free parameters (7)
  • number_of_document_examples = 4 (2 for FUNSD when tokens exceed limit)
    Selected based on CORD test performance; tested 1-5 in Appendix B.
  • number_of_layout_examples = 4
    Same as document examples; not individually ablated in main text.
  • number_of_entity_examples = 4
    Ablated on FUNSD: 2 vs 4 (Table A7); 4 chosen.
  • layout_similarity_metric = MSE
    Chosen by comparing MSE, cosine, SSIM on CORD (Table A3).
  • resize_method = LANCZOS interpolation and binarization
    Chosen by comparing five resize methods on CORD (Table A4).
  • entity_representation_format = text: "...", Box: [x1,y1,x2,y2], entity: ...
    Chosen on FUNSD (Table A6).
  • example_order = descending similarity for layout and text
    Chosen on CORD (Table 4).
assumptions (5)
  • domain assumption Sentence-BERT embeddings capture semantic similarity sufficient for selecting helpful ICL examples
    Used throughout Section 3.3-3.4; no evidence that other embeddings would work as well.
  • domain assumption MSE between resized binary layout images measures layout similarity relevant to label positions
    Section 3.5; the paper compares with cosine and SSIM but does not validate against human judgment.
  • domain assumption The LLMs (ChatGLM3, GPT-3.5, GPT-4o) follow the constructed prompts reliably
    Section 4.1; no analysis of failure modes or prompt sensitivity.
  • domain assumption OCR outputs T and B are accurate and available for all datasets
    Section 3.1; the exact OCR engine is not specified, so this is an unverified precondition.
  • standard math Wilcoxon signed-rank test assumptions hold for paired F1 samples
    Appendix C; standard.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAIL: Sample-Centric In-Context Learning for Document Information Extraction." pith.science (2026). https://pith.science/paper/OVS3DCC6

@misc{pith2026241217092,
  author       = {Pith},
  title        = {Pith review of: SAIL: Sample-Centric In-Context Learning for Document Information Extraction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OVS3DCC6}},
  note         = {Machine review of arXiv:2412.17092}
}
read the original abstract

Document Information Extraction (DIE) aims to extract structured information from Visually Rich Documents (VRDs). Previous full-training approaches have demonstrated strong performance but may struggle with generalization to unseen data. In contrast, training-free methods leverage powerful pre-trained models like Large Language Models (LLMs) to address various downstream tasks with only a few examples. Nonetheless, training-free methods for DIE encounter two primary challenges: (1) understanding the complex relationship between layout and textual elements in VRDs, and (2) providing accurate guidance to pre-trained models. To address these challenges, we propose Sample-centric In-context Learning (SAIL) for DIE. SAIL introduces a fine-grained entity-level textual similarity to facilitate in-depth text analysis by LLMs and incorporates layout similarity to enhance the analysis of layouts in VRDs. Additionally, SAIL formulates a unified In-Context Learning (ICL) prompt template for various sample-centric examples, enabling tailored prompts that deliver precise guidance to pre-trained models for each sample. Extensive experiments on FUNSD, CORD, and SROIE benchmarks with various base models (e.g., LLMs) indicate that our method outperforms training-free baselines, even closer to the full-training methods. The results show the superiority and generalization of our method.

Figures

Figures reproduced from arXiv: 2412.17092 by the authors.

Figure 1
Figure 1. For the (a) test sample from the CORD dataset [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Illustration of SAIL framework, including extracting texts T and boxes B from document images, encoding them separately, selecting textually similar entities, layout similar documents, and textually similar documents for each test sample, constructing sample-centric prompts using diverse examples, and generating predicted labels. goal is to label all entities within I. First, entity texts T = {t1, t2, ..., tne } and… view at source ↗
Figure 3
Figure 3. Illustration of layout similarity evaluation, in￾cluding drawing boxes onto a blank image, cropping and re￾sizing to form layout image, and comparing layout images. is notably low. To facilitate LLMs in generating text with more relevant examples for learning, we propose entity-level text similarity examples, as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Case study on performance comparison of (a) ICL￾D3IE and (b) our SAIL. ICL-D3IE wrongly predicts the three green boxes on the left as “answer”. In contrast, our proposed SAIL correctly predicts them as “question”. # Similar FUNSD CORD SROIE Text-Doc. Layout Text-Ent. 0…
Figure 6
Figure 6. Figure 6: (a) Case study on comparison of (a1) without and (a2) with layout similar examples. Adding layout similar examples helps accurately distinguish between the two “13.000”. (b) Case study on comparison of (b1) without and (b2) with entity-level text similar examples. The …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 40 canonical work pages

  1. [1]

    Abramovich, O.; Nayman, N.; Fogel, S.; Lavi, I.; Litman, R.; Tsiper, S.; Tichauer, R.; Appalaraju, S.; Mazor, S.; and Manmatha, R. 2024. VisFocus: Prompt-guided vision encoders for ocr-free dense document understanding. In ECCV

    work page 2024

  2. [2]

    D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; et al

    Brown, T.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J. D.; Dhariwal, P.; Neelakantan, A.; Shyam, P.; Sastry, G.; Askell, A.; et al. 2020. Language models are few-shot learners. In NeurIPS

    work page 2020

  3. [3]

    Cai, C.; Wang, Q.; Liang, B.; Qin, B.; Yang, M.; Wong, K.-F.; and Xu, R. 2023. In-context learning for few-shot multimodal named entity recognition. In Findings of the Association for Computational Linguistics: EMNLP

    work page 2023

  4. [4]

    Da, C.; Luo, C.; Zheng, Q.; and Yao, C. 2023. Vision grid transformer for document layout analysis. In ICCV

    work page 2023

  5. [5]

    I.; and Reisswig, C

    Denk, T. I.; and Reisswig, C. 2019. Bertgrid: Contextualized embedding for 2d document representation and understanding. In Workshop on Document Intelligence at NeurIPS

    work page 2019

  6. [6]

    Devlin, J.; Chang, M.-W.; Lee, K.; and Toutanova, K. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL-HLT

    work page 2019

  7. [7]

    Fujitake, M. 2024. LayoutLLM: Large language model instruction tuning for visually rich document understanding. In LREC-COLING

    work page 2024

  8. [8]

    Gu, Z.; Meng, C.; Wang, K.; Lan, J.; Wang, W.; Gu, M.; and Zhang, L. 2022. Xylayoutlm: Towards layout-aware multimodal networks for visually-rich document understanding. In CVPR

    work page 2022

Show all 49 references
  1. [9]

    He, J.; Wang, L.; Hu, Y.; Liu, N.; Liu, H.; Xu, X.; and Shen, H. T. 2023. Icl-d3ie: In-context learning with diverse demonstrations updating for document information extraction. In ICCV

    2023

  2. [10]

    Hong, T.; Kim, D.; Ji, M.; Hwang, W.; Nam, D.; and Park, S. 2022. Bros: A pre-trained language model focusing on text and layout for better key information extraction from documents. In AAAI

    2022

  3. [11]

    Hu, A.; Xu, H.; Ye, J.; Yan, M.; Zhang, L.; Zhang, B.; Li, C.; Zhang, J.; Jin, Q.; Huang, F.; et al. 2024. Mplug-docowl 1.5: Unified structure learning for ocr-free document understanding. arXiv preprint arXiv:2403.12895

    2024 arXiv

  4. [12]

    A.; and Ostendorf, M

    Hu, Y.; Lee, C.-H.; Xie, T.; Yu, T.; Smith, N. A.; and Ostendorf, M. 2022. In-context learning for few-shot dialogue state tracking. In Findings of the Association for Computational Linguistics: EMNLP

    2022

  5. [13]

    Huang, Y.; Lv, T.; Cui, L.; Lu, Y.; and Wei, F. 2022. Layoutlmv3: Pre-training for document ai with unified text and image masking. In ACM MM

    2022

  6. [14]

    Huang, Z.; Chen, K.; He, J.; Bai, X.; Karatzas, D.; Lu, S.; and Jawahar, C. 2019. Icdar2019 competition on scanned receipt ocr and information extraction. In ICDAR

    2019

  7. [15]

    K.; and Thiran, J.-P

    Jaume, G.; Ekenel, H. K.; and Thiran, J.-P. 2019. Funsd: A dataset for form understanding in noisy scanned documents. In ICDARW

    2019

  8. [16]

    R.; Reisswig, C.; Guder, C.; Brarda, S.; Bickel, S.; H \"o hne, J.; and Faddoul, J

    Katti, A. R.; Reisswig, C.; Guder, C.; Brarda, S.; Bickel, S.; H \"o hne, J.; and Faddoul, J. B. 2018. Chargrid: Towards understanding 2d documents. In EMNLP

    2018

  9. [17]

    Kerroumi, M.; Sayem, O.; and Shabou, A. 2021. VisualWordGrid: Information extraction from scanned documents using a multimodal approach. In ICDAR

    2021

  10. [18]

    Kim, G.; Hong, T.; Yim, M.; Nam, J.; Park, J.; Yim, J.; Hwang, W.; Yun, S.; Han, D.; and Park, S. 2022. Ocr-free document understanding transformer. In ECCV

    2022

  11. [19]

    Li, C.; Bi, B.; Yan, M.; Wang, W.; Huang, S.; Huang, F.; and Si, L. 2021. StructuralLM: Structural pre-training for form understanding. In ACL-IJCNLP

    2021

  12. [20]

    Li, X.; Wu, Y.; Jiang, X.; Guo, Z.; Gong, M.; Cao, H.; Liu, Y.; Jiang, D.; and Sun, X. 2024. Enhancing visual document understanding with contrastive learning in large visual-language models. In CVPR

    2024

  13. [21]

    Liu, H.; Li, C.; Li, Y.; and Lee, Y. J. 2024 a . Improved baselines with visual instruction tuning. In CVPR

    2024

  14. [22]

    B.; Carin, L.; and Chen, W

    Liu, J.; Shen, D.; Zhang, Y.; Dolan, W. B.; Carin, L.; and Chen, W. 2022. What makes good in-context examples for GPT-3? In DeeLIO: The 3rd Workshop on Knowledge Extraction and Integration for Deep Learning Architectures

    2022

  15. [23]

    Liu, W.; Lei, F.; Luo, T.; Lei, J.; He, S.; Zhao, J.; and Liu, K. 2023. MMHQA-ICL: Multimodal in-context learning for hybrid question answering over text, tables and images. arXiv preprint arXiv:2309.04790

    2023 arXiv

  16. [24]

    Liu, X.; Gao, F.; Zhang, Q.; and Zhao, H. 2019. Graph convolution for multimodal information extraction from visually rich documents. In NAACL-HLT

    2019

  17. [25]

    Liu, Y.; Yang, B.; Liu, Q.; Li, Z.; Ma, Z.; Zhang, S.; and Bai, X. 2024 b . Textmonkey: An ocr-free large multimodal model for understanding document. arXiv preprint arXiv:2403.04473

    2024 arXiv

  18. [26]

    Lu, J.; Yu, H.; Wang, Y.; Ye, Y.; Tang, J.; Yang, Z.; Wu, B.; Liu, Q.; Feng, H.; Wang, H.; et al. 2024. A bounding box is worth one token: Interleaving layout and text in a large language model for document understanding. arXiv preprint arXiv:2407.01976

    2024 arXiv

  19. [27]

    Luo, C.; Shen, Y.; Zhu, Z.; Zheng, Q.; Yu, Z.; and Yao, C. 2024. LayoutLLM: Layout instruction tuning with large language models for document understanding. In CVPR

    2024

  20. [28]

    Mao, Z.; Bai, H.; Hou, L.; Wei, J.; Jiang, X.; Liu, Q.; and Wong, K.-F. 2024. Visually guided generative text-layout pre-training for document intelligence. In NAACL-HLT

    2024

  21. [29]

    Meade, N.; Gella, S.; Hazarika, D.; Gupta, P.; Jin, D.; Reddy, S.; Liu, Y.; and Hakkani-T \"u r, D. 2023. Using in-context learning to improve dialogue safety. In Findings of the Association for Computational Linguistics: EMNLP

    2023

  22. [30]

    OpenAI. 2023 a . gpt-3.5-turbo

    2023

  23. [31]

    OpenAI. 2023 b . GPT-4o System Card

    2023

  24. [32]

    Park, S.; Shin, S.; Lee, B.; Lee, J.; Surh, J.; Seo, M.; and Lee, H. 2019. CORD: A consolidated receipt dataset for post-ocr parsing. In NeurIPS Workshop

    2019

  25. [33]

    S.; Wang, Z.; Mu, J.; Zhang, H.; and Hua, N

    Perot, V.; Kang, K.; Luisier, F.; Su, G.; Sun, X.; Boppana, R. S.; Wang, Z.; Mu, J.; Zhang, H.; and Hua, N. 2023. Lmdx: Language model-based document information extraction and localization. arXiv preprint arXiv:2309.10952

    2023 arXiv

  26. [34]

    Qian, Y.; Santus, E.; Jin, Z.; Guo, J.; and Barzilay, R. 2019. Graphie: A graph-based framework for information extraction. In NAACL-HLT

    2019

  27. [35]

    Reimers, N.; and Gurevych, I. 2019. Sentence-BERT: Sentence embeddings using siamese BERT-networks. In EMNLP-IJCNLP

    2019

  28. [36]

    Tang, G.; Xie, L.; Jin, L.; Wang, J.; Chen, J.; Xu, Z.; Wang, Q.; Wu, Y.; and Li, H. 2021. Matchvie: Exploiting match relevancy between entities for visual information extraction. arXiv preprint arXiv:2106.12940

    2021 arXiv

  29. [37]

    THUDM. 2023. Chatglm3

    2023

  30. [38]

    N.; Kaiser, .; and Polosukhin, I

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, .; and Polosukhin, I. 2017. Attention is all you need. In NeurIPS

    2017

  31. [39]

    Wang, D.; Raman, N.; Sibue, M.; Ma, Z.; Babkin, P.; Kaur, S.; Pei, Y.; Nourbakhsh, A.; and Liu, X. 2023 a . DocLLM: A layout-aware generative language model for multimodal document understanding. arXiv preprint arXiv:2401.00908

    2023 arXiv

  32. [40]

    Wang, J.; Jin, L.; and Ding, K. 2022. Lilt: A simple yet effective language-independent layout transformer for structured document understanding. In ACL

    2022

  33. [41]

    Wang, J.; Le, X.; Peng, X.; and Chen, C. 2023 b . Adaptive hinge balance loss for document-level relation extraction. In Findings of the Association for Computational Linguistics: EMNLP

    2023

  34. [42]

    Wang, J.; Liu, C.; Jin, L.; Tang, G.; Zhang, J.; Zhang, S.; Wang, Q.; Wu, Y.; and Cai, M. 2021. Towards robust visual information extraction in real world: new dataset and novel solution. In AAAI

    2021

  35. [43]

    Wang, L.; Hu, Y.; He, J.; Xu, X.; Liu, N.; Liu, H.; and Shen, H. T. 2024. T-sciq: Teaching multimodal chain-of-thought reasoning via large language model signals for science question answering. In AAAI

    2024

  36. [44]

    Xu, Y.; Li, M.; Cui, L.; Huang, S.; Wei, F.; and Zhou, M. 2020. Layoutlm: Pre-training of text and layout for document image understanding. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining

    2020

  37. [45]

    Xu, Y.; Xu, Y.; Lv, T.; Cui, L.; Wei, F.; Wang, G.; Lu, Y.; Florencio, D.; Zhang, C.; Che, W.; et al. 2021. Layoutlmv2: Multi-modal pre-training for visually-rich document understanding. In ACL-IJCNLP

    2021

  38. [46]

    Yang, Z.; Gan, Z.; Wang, J.; Hu, X.; Lu, Y.; Liu, Z.; and Wang, L. 2022. An empirical study of gpt-3 for few-shot knowledge-based vqa. In AAAI

    2022

  39. [47]

    Zhao, X.; Niu, E.; Wu, Z.; and Wang, X. 2019. Cutie: Learning to understand documents with convolutional universal text information extractor. arXiv preprint arXiv:1903.12363

    2019 arXiv

  40. [48]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  41. [49]

    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 gl...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.