Pith. sign in

REVIEW 3 major objections 6 minor 42 references

Comparison of Feature Learning Methods for Metadata Extraction from PDF Scholarly Documents

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that TextMap-Word2Vec, a layout-aware approach that fuses word embeddings with page geometry, achieves the best metadata extraction score on a new 50,000-document benchmark—though its own results table reports a…

desk verdict TextMap is a real architectural idea, but the paper's empirical claims are undercut by internally inconsistent F1 numbers and a headline result contradicted by its own Table III. read the letter →

arxiv 2501.05082 v1 pith:4CBHPWMY submitted 2025-01-09 cs.IR cs.CLcs.DLcs.LG

Zeyd Boukhers , Cong Yang This is my paper · ORCID
classification cs.IRcs.CLcs.DLcs.LG
keywords metadataextractiondocumentprocessingneuralnetworksnaturallanguagecomputervisionmultimodalapproachesscientificdocumentsTextMap
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

This paper evaluates ways to pull metadata—title, authors, abstract, DOI, and similar fields—out of the first page of a scholarly PDF, focusing on documents with highly varied templates, such as those from small and mid-sized German social-science publishers. Its central proposal is TextMap, which maps text embeddings onto the spatial coordinates of the page, combines them with a grayscale layout image through attention, and runs a Fast R-CNN detector to label regions. The paper reports that TextMap with Word2Vec embeddings reaches the highest F1-score of 0.913 on its new SSOAR-MVD dataset. However, Table III in the same paper lists the Vision-Language method at 0.92, which is larger than 0.913, so the "highest" claim is contradicted by the paper's own numbers. If the approach holds up, it offers a cheaper, layout-aware alternative to full multimodal pipelines for making under-indexed documents findable.

What carries the argument

The central mechanism is the TextMap interpolation–fusion pipeline: (1) a spatial stream converts page $P$ into grayscale $G = \phi(P) \in \mathbb{R}^{H \times W}$; (2) a semantic stream embeds tokens with Word2Vec or blocks with BERT and maps each embedding into the bounding-box coordinates of the region where that text appears, forming $T_m \in \mathbb{R}^{H \times W \times d}$; (3) convolutional layers bring both streams to a common channel dimension, and multi-head attention $M = \mathrm{Attention}(F_{spatial}, F_{semantic})V$ fuses them while preserving spatial correspondence; (4) a Fast R-CNN head refines boxes and assigns metadata labels. A joint loss combines a semantic term, a spatial smoothness term that encourages neighboring same-label regions to have similar features, and a cross-modal term. The load-bearing innovation is that the text map keeps semantic information tied to page coordinates, so the detector can exploit layout even when templates vary.

What would settle it

Re-annotate a random sample of several hundred SSOAR-MVD pages with human labelers marking title, authors, abstract, DOI, and the other fields, then recompute F1 for TextMap-Word2Vec and Vision-Language on the same sample; if TextMap-Word2Vec no longer beats Vision-Language or drops below 0.913, the paper's central claim fails. A simpler check is to read Table III directly: it already lists Vision-Language at 0.92, above the claimed 0.913, so the "highest" assertion is internally falsified unless the authors meant "best among newly proposed methods."

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that jointly optimizing spatial and semantic information improves metadata extraction from template-varied PDFs. The TextMap method represents the first page as a grayscale image in one stream and as token or block embeddings in another, then interpolates those embeddings into the regions where the text appears, producing a "text map" with the same spatial dimensions as the page; convolutions on both streams are fused by multi-head attention and fed to a Fast R-CNN for label and bounding-box prediction. The authors state that the Word2Vec variant achieves the best overall F1-score, 0.913, on SSOAR-MVD and 0.901 on S-PMRD. The paper's own Table III, however, shows Vision-Language at 0.92 on SSOAR-MVD, so the asserted "highest" result does not match the table; a sympathetic reading is that the authors intended to claim TextMap-Word2Vec is the best among the newly introduced methods, not among all compared methods. The paper also introduces two datasets, SSOAR-MVD and S-PMRD, and reports per-category scores plus training and inference times for nine methods.

Load-bearing premise

The load-bearing assumption is that the automatic similarity-based labeling of SSOAR-MVD is correct ground truth: blocks are assigned to metadata classes only when they have near-perfect similarity to the repository's textual metadata, with the threshold unspecified and no reported manual validation, so every precision, recall, and F1 number inherits whatever noise or bias those labels contain.

Editorial extensions

If this is right

  • If TextMap's reported scores are taken at face value, layout-aware fusion of embeddings with page geometry reaches about 0.91 F1 on both benchmarks, competitive with a much more expensive multimodal model.
  • The Word2Vec variant's cost profile (about 92 hours of training, 0.4 seconds per page at inference) makes it a practical choice when GPU time or annotated data is limited.
  • The two new datasets, SSOAR-MVD and S-PMRD, give the community a template-diverse German social-science benchmark and an authentic S2ORC-based PDF benchmark for future comparisons.
  • The per-category scores suggest structured fields such as DOI and Email are easiest, while Affiliation and Address remain the weak spots across nearly all methods.
  • The internal contradiction between the claimed highest F1 of 0.913 and Table III's 0.92 for Vision-Language would need to be corrected in a revision before the headline result can be accepted.

Reading between the lines

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

  • Editorial inference: if the 0.92 Vision-Language number is the true best, then the paper's conclusion shifts from "TextMap is best" to "multimodal fusion wins but costs about twice as much to train and nearly nine times as long to run," which is still a useful engineering trade-off.
  • Editorial inference: because the SSOAR-MVD labels were generated automatically by similarity matching with no reported manual validation, the absolute F1 values in Tables III through XII may be optimistic; a human-annotated sample would be needed to confirm the ranking.
  • Editorial inference: the TextMap idea of interpolating embeddings into dense spatial maps is not limited to metadata; it could be transferred to table-structure recognition, form understanding, or any document task where layout and semantics both matter.
  • Editorial inference: a direct, testable follow-up is to run TextMap-Word2Vec and Vision-Language on the same 50,000 documents with human-labeled boxes and compare F1; whichever wins would resolve the table-vs-text discrepancy.
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

3 major / 6 minor

Summary. The manuscript compares nine feature learning and prediction methods for extracting metadata from the first page of scholarly PDFs: CRF, BiLSTM, BiLSTM-CRF, GROBID, a Mask/Fast R-CNN vision model, a vision-language multimodal model, and a new TextMap approach instantiated with BERT, Word2Vec, and Char2Vec embeddings. The comparison is carried out on two datasets: SSOAR-MVD, a 50,000-document set synthesized from SSOAR metadata and templates, and S-PMRD, a curated subset of S2ORC. The paper reports token-level precision, recall, and F1 for each method, and claims in Section V.A that TextMap-Word2Vec achieves the highest F1-score of 0.913 on SSOAR-MVD. It also reports training and inference times and closes with a discussion of limitations.

Significance. If the reported results were reliable, the paper would offer a practical comparison of established and new extraction methods under high template variance, and the two datasets could serve as useful benchmarks. The proposed TextMap architecture, which fuses a grayscale page image with spatially interpolated text embeddings, is a plausible contribution. However, the central empirical claim is contradicted by the paper's own Tables III and IV, and the F1 scores in several tables are not arithmetically consistent with the stated precision and recall values. In addition, the SSOAR-MVD ground truth is generated by an unvalidated automatic labeling procedure. These issues undermine the headline comparison and the reliability of the datasets as presented, so the current contribution cannot be assessed as a trustworthy empirical study.

major comments (3)
  1. [Section V.A, Table III] The claim that 'the proposed TextMap-Word2Vec method achieves the highest F1-score of 0.913' is directly contradicted by Table III, which reports Vision-Language with an F1-score of 0.920 and TextMap-Word2Vec with 0.913. The same ordering appears in Table IV for S-PMRD, where Vision-Language (0.903) edges out TextMap-Word2Vec (0.901). No qualification is given that these F1 values are computed under different protocols, so the paper's headline result is unsupported by its own data.
  2. [Tables V-XII] Many reported F1 values are not derivable from the listed precision and recall. For example, Table VIII (GROBID) gives Title precision 0.764 and recall 0.667, from which F1 = 2*0.764*0.667/(0.764+0.667) = 0.712, yet the table reports 0.951; the Abstract row (P=0.84, R=0.79) yields F1=0.814, not 0.935; the Email row (P=0.91, R=0.812) yields F1=0.858, not 0.893. These are not isolated typos but a pattern of systematic inconsistency in the metric tables, so the comparisons cannot be trusted without corrected computations or access to the underlying predictions. Footnote 4 states that the code 'will be released upon publication,' but no artifact is available to resolve the discrepancy.
  3. [Section IV.A.1] The SSOAR-MVD ground truth is created by an automatic labeling procedure that assigns a text block to a metadata class when it has 'near-perfect similarity' to the corresponding SSOAR metadata value. The similarity measure and threshold are not specified, and no manual validation, inter-annotator agreement, or error analysis is reported. Every precision, recall, and F1 value in the paper depends on this unverified labeling; if the labels are noisy or biased, all cross-method comparisons are affected. The authors should define the similarity criterion, report its threshold, and provide evidence of labeling quality.
minor comments (6)
  1. [Section III.E] The subsection is titled 'Fast-RCNN' and the result tables refer to 'Fast-RCNN', but the described architecture is Mask R-CNN with a mask prediction branch and RoIAlign; this naming inconsistency confuses the method description.
  2. [Table XI] The header 'Macro Avrerage' contains a typo; it should read 'Macro Average'.
  3. [Footnote 4] The footnote text 'Willbereleaseduponpublicaiton' is garbled; it should read 'Will be released upon publication.'
  4. [References] Several references are duplicated: [3] and [24] are the same MexPub paper, [25] and [33] are the same Mask R-CNN paper, [26] and [34] are the same ResNeXt paper, and [9] and [27] are the same lecture-video metadata paper; these duplicates should be consolidated.
  5. [Table XIII] Some entries report standard deviations while others do not (e.g., '0.4 seconds' for TextMap-Word2Vec), and units are sometimes omitted after numeric values (e.g., '92 ± 6.2'); this makes the computational cost comparison difficult to interpret.
  6. [Section I] The introduction states that 'Section IV presents the dataset and experimental results,' but the results appear in Section V; the section cross-references should be corrected.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the paper is an empirical benchmark comparison, and none of its reported results reduce to their inputs by construction; the internal F1 inconsistencies are correctness/verifiability problems, not circularity.

full rationale

The paper's contribution is a comparative evaluation, not a derivation; the TextMap framework (Section III-G, Eqs. 9-24) is specified independently of the metrics and trained on the same datasets as the baselines. The two self-citations [3], [4] describe the authors' own Fast-RCNN and Vision-Language baselines and are not used to justify a premise or forbid alternatives, so they are not load-bearing. The automatic labeling of SSOAR-MVD (Section IV.A.1) derives bounding-box ground truth from SSOAR metadata by text similarity; this is a standard supervised labeling procedure, not a feedback loop in which model outputs define the prediction target, so no step is self-definitional. The paper does contain serious correctness and reproducibility flaws that are outside the circularity category: Table III reports Vision-Language at 0.92 while the text claims TextMap-Word2Vec has the highest F1 of 0.913, and several per-category F1 values are arithmetically impossible relative to their reported precision and recall (e.g., Table V Title P=0.568, R=0.350, F1=0.433; Table VIII Title P=0.764, R=0.667, F1=0.951). Footnote 4 states code 'will be released upon publication,' so no artifact is currently available to resolve these anomalies. These problems undermine the reported comparison, but they are empirical/verification issues rather than circular reductions of equations to fitted values. Therefore no circular step is established and the score is 0.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on assumptions about dataset labeling quality and the first-page localization of metadata. The TextMap model introduces no new physical or conceptual entities; its components are standard. The auto-labeling assumption is the most fragile, since all results depend on it.

free parameters (4)
  • Learning rate for Fast-RCNN fine-tuning = 2.5e-3
    Set by hand; no sensitivity analysis reported (Section III-E).
  • Number of frozen layers and iterations for Fast-RCNN = 2 frozen layers, 15k iterations
    Chosen based on experiments on 103 random samples (Section III-E).
  • Similarity threshold for auto-labeling = unspecified ('near-perfect similarity')
    The threshold for assigning bounding boxes to metadata classes is not quantified (Section IV.A.1).
  • Various neural network hyperparameters = hidden sizes 112/115/256, layer counts, dropout, etc.
    Chosen by hand; no grid search or sensitivity analysis reported (Sections III-B, III-C, III-F).
assumptions (3)
  • domain assumption Metadata for a scientific document is fully contained on the first page.
    Stated in Section III: 'we assume that metadata may only be present on the first page of a PDF document'.
  • domain assumption String similarity between extracted text blocks and repository metadata yields correct ground-truth labels.
    The SSOAR-MVD and S-PMRD labels are generated automatically by matching text blocks to metadata with 'near-perfect similarity' or exact/fuzzy matching (Section IV.A). No manual validation is described.
  • ad hoc to paper A grayscale rendering of the page plus interpolated text embeddings can be fused via attention to predict metadata classes.
    This is the core modeling assumption of TextMap (Section III-G), not derived from established theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Comparison of Feature Learning Methods for Metadata Extraction from PDF Scholarly Documents." pith.science (2026). https://pith.science/paper/4CBHPWMY

@misc{pith2026250105082,
  author       = {Pith},
  title        = {Pith review of: Comparison of Feature Learning Methods for Metadata Extraction from PDF Scholarly Documents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4CBHPWMY}},
  note         = {Machine review of arXiv:2501.05082}
}
read the original abstract

The availability of metadata for scientific documents is pivotal in propelling scientific knowledge forward and for adhering to the FAIR principles (i.e. Findability, Accessibility, Interoperability, and Reusability) of research findings. However, the lack of sufficient metadata in published documents, particularly those from smaller and mid-sized publishers, hinders their accessibility. This issue is widespread in some disciplines, such as the German Social Sciences, where publications often employ diverse templates. To address this challenge, our study evaluates various feature learning and prediction methods, including natural language processing (NLP), computer vision (CV), and multimodal approaches, for extracting metadata from documents with high template variance. We aim to improve the accessibility of scientific documents and facilitate their wider use. To support our comparison of these methods, we provide comprehensive experimental results, analyzing their accuracy and efficiency in extracting metadata. Additionally, we provide valuable insights into the strengths and weaknesses of various feature learning and prediction methods, which can guide future research in this field.

Figures

Figures reproduced from arXiv: 2501.05082 by the authors.

Figure 1
Figure 1. Schematic representation of the two-layer Conditional Random Field [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Diagram of the Bi-Directional LSTM network architecture for [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Overview of the GROBID Framework for structured metadata [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Mask R-CNN architecture employed for metadata extraction from PDF pages. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Multimodal extraction approach, where (a) refers to the model described in section III-E and (b) refers to the model described in section III-B ELMO [40] embedding results, derived from a model trained on German documents. Following the two LSTM layers, a fully connect…
Figure 7
Figure 7. Figure 7: Overview of the TextMap approach a) Embedding Interpolation: For each region Ri , the embedding is directly mapped into the spatial coordinates of that region to create the text map Tm. We perform a straightforward mapping to ensure that each spatial location in the te…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 28 canonical work pages

  1. [1]

    Openaire research graph dump,

    P. Manghi, C. Atzori, A. Bardi, M. Baglioni, J. Schirrwagen, H. Dim- itropoulos, S. La Bruzzo, I. Foufoulas, A. Mannocci, M. Horst et al. , “Openaire research graph dump,” 2022

    work page 2022

  2. [2]

    Open research knowledge graph: next generation infrastructure for semantic scholarly knowledge,

    M. Y . Jaradeh, A. Oelen, K. E. Farfar, M. Prinz, J. D’Souza, G. Kismih´ok, M. Stocker, and S. Auer, “Open research knowledge graph: next generation infrastructure for semantic scholarly knowledge,” in Proceedings of the 10th international conference on knowledge capture , 2019, pp. 243–246

    work page 2019

  3. [3]

    Mexpub: Deep transfer learning for metadata extraction from german publications,

    Z. Boukhers, N. Beili, T. Hartmann, P. Goswami, and M. A. Zafar, “Mexpub: Deep transfer learning for metadata extraction from german publications,” in 2021 ACM/IEEE Joint Conference on Digital Libraries (JCDL). IEEE, 2021, pp. 250–253

    work page 2021

  4. [4]

    Vision and natural language for meta- data extraction from scientific pdf documents: a multimodal approach,

    Z. Boukhers and A. Bouabdallah, “Vision and natural language for meta- data extraction from scientific pdf documents: a multimodal approach,” in Proceedings of the 22nd ACM/IEEE Joint Conference on Digital Libraries, 2022, pp. 1–5

    work page 2022

  5. [5]

    Clinical information extrac- tion applications: a literature review,

    Y . Wang, L. Wang, M. Rastegar-Mojarad, S. Moon, F. Shen, N. Afzal, S. Liu, Y . Zeng, S. Mehrabi, S. Sohnet al., “Clinical information extrac- tion applications: a literature review,” Journal of biomedical informatics, vol. 77, pp. 34–49, 2018

    work page 2018

  6. [6]

    Advances in natural language processing,

    J. Hirschberg and C. D. Manning, “Advances in natural language processing,” Science, vol. 349, no. 6245, pp. 261–266, 2015

    2015

  7. [7]

    An efficient framework for metadata extraction over scholarly documents using ensemble cnn and bilstm technique,

    P. R. Nayaka and R. Ranjan, “An efficient framework for metadata extraction over scholarly documents using ensemble cnn and bilstm technique,” in 2023 2nd International Conference for Innovation in Technology (INOCON). IEEE, 2023, pp. 1–9

    work page 2023

  8. [8]

    Computer vision and machine learning approaches for metadata enrichment to improve searchability of historical newspaper collections,

    D. Ali, K. Milleville, S. Verstockt, N. Van de Weghe, S. Chambers, and J. M. Birkholz, “Computer vision and machine learning approaches for metadata enrichment to improve searchability of historical newspaper collections,” Journal of Documentation , 2023

    work page 2023

Show all 42 references
  1. [9]

    A multimodal approach for extracting content descriptive metadata from lecture videos,

    V . Balasubramanian, S. G. Doraisamy, and N. K. Kanakarajan, “A multimodal approach for extracting content descriptive metadata from lecture videos,” Journal of Intelligent Information Systems , vol. 46, pp. 121–145, 2016

    2016

  2. [10]

    Reference metadata extraction using a hierarchical knowledge representation framework,

    M.-Y . Day, R. T.-H. Tsai, C.-L. Sung, C.-C. Hsieh, C.-W. Lee, S.-H. Wu, K.-P. Wu, C.-S. Ong, and W.-L. Hsu, “Reference metadata extraction using a hierarchical knowledge representation framework,” Decision Support Systems, vol. 43, no. 1, pp. 152–167, 2007. [Online]. Availabl...

    2007

  3. [11]

    A unified framework for automatic metadata extraction from electronic document,

    A. Kawtrakul and C. Yingsaeree, “A unified framework for automatic metadata extraction from electronic document,” in Proceedings of The International Advanced Digital Library Conference. Nagoya, Japan , 2005

    2005

  4. [12]

    Rule-based word clustering for document metadata extraction,

    H. Han, E. Manavoglu, H. Zha, K. Tsioutsiouliklis, C. L. Giles, and X. Zhang, “Rule-based word clustering for document metadata extraction,” in Proceedings of the 2005 ACM Symposium on Applied Computing, ser. SAC ’05. New York, NY , USA: Association for Computing Machinery, 20...

    2005

  5. [13]

    Citeseerx: an architecture and web service design for an academic document search engine,

    H. Li, I. Councill, W.-C. Lee, and C. L. Giles, “Citeseerx: an architecture and web service design for an academic document search engine,” in Proceedings of the 15th international conference on World Wide Web , 2006, pp. 883–884

    2006

  6. [14]

    Learning hidden markov model structure for information extraction,

    K. Seymore, A. Mccallum, and R. Rosenfeld, “Learning hidden markov model structure for information extraction,” in In AAAI 99 Workshop on Machine Learning for Information Extraction , 1999, pp. 37–42

    1999

  7. [15]

    Information extraction from research papers using conditional random fields,

    F. Peng and A. McCallum, “Information extraction from research papers using conditional random fields,” Inf. Process. Manage. , vol. 42, no. 4, p. 963–979, 2006. [Online]. Available: https: //doi.org/10.1016/j.ipm.2005.09.002

    2006 doi

  8. [16]

    Auto- matic document metadata extraction using support vector machines,

    H. Han, C. Giles, E. Manavoglu, H. Zha, Z. Zhang, and E. Fox, “Auto- matic document metadata extraction using support vector machines,” in 2003 Joint Conference on Digital Libraries, 2003. Proceedings. , 2003, pp. 37–48

    2003

  9. [17]

    Bidirectional lstm-crf models for sequence tagging,

    Z. Huang, W. Xu, and K. Yu, “Bidirectional lstm-crf models for sequence tagging,” CoRR, vol. abs/1508.01991, 2015. [Online]. Available: http://arxiv.org/abs/1508.01991

    2015 arXiv

  10. [18]

    Named entity recognition with bidirectional lstm-cnns,

    J. P. C. Chiu and E. Nichols, “Named entity recognition with bidirectional lstm-cnns,” CoRR, vol. abs/1511.08308, 2015. [Online]. Available: http://arxiv.org/abs/1511.08308

    2015 arXiv

  11. [19]

    Citation metadata extraction via deep neural network-based segment sequence labeling,

    D. An, L. Gao, Z. Jiang, R. Liu, and Z. Tang, “Citation metadata extraction via deep neural network-based segment sequence labeling,” in Proceedings of the 2017 ACM on Conference on Information and Knowledge Management , ser. CIKM ’17. New York, NY , USA: Association for Compu...

    2017

  12. [20]

    Parscit: an open-source crf reference string parsing package,

    I. G. Councill, C. L. Giles, and M.-Y . Kan, “Parscit: an open-source crf reference string parsing package,” LREC, Vol. 8., p. 661–667, 2008

    2008

  13. [21]

    Bibpro: A citation parser based on sequence alignment,

    C.-C. Chen, K.-H. Yang, C.-L. Chen, and J.-M. Ho, “Bibpro: A citation parser based on sequence alignment,” IEEE Transactions on Knowledge and Data Engineering 24, 2 (2012) , p. 236–250, 2012

    2012

  14. [22]

    A new dataset for fine-grained citation field extraction,

    S. Anzaroot and A. Mccallum, “A new dataset for fine-grained citation field extraction,” ICML Workshop on Peer Reviewing and Publishing Models., 2013

    2013

  15. [23]

    Deeppdf: A deep learning approach to extracting text from pdfs,

    C. G. Stahl, S. R. Young, D. Herrmannova, R. M. Patton, and J. C. Wells, “Deeppdf: A deep learning approach to extracting text from pdfs,” 2018. [Online]. Available: https://www.osti.gov/biblio/1460210

    2018

  16. [24]

    Mexpub: Deep transfer learning for metadata extraction from german publications,

    Z. Boukhers, N. Beili, T. Hartmann, P. Goswami, and M. A. Zafar, “Mexpub: Deep transfer learning for metadata extraction from german publications,” in 2021 ACM/IEEE Joint Conference on Digital Libraries (JCDL). IEEE, 2021

    2021

  17. [25]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” in 2017 IEEE International Conference on Computer Vision (ICCV) , 2017, pp. 2980–2988

    2017

  18. [26]

    Aggregated residual transformations for deep neural networks,

    S. Xie, R. Girshick, P. Doll ´ar, Z. Tu, and K. He, “Aggregated residual transformations for deep neural networks,” arXiv preprint arXiv:1611.05431, 2016

    2016 arXiv

  19. [27]

    A multimodal approach for extracting content descriptive metadata from 14 lecture videos,

    V . Balasubramanian, S. G. Doraisamy, and N. K. Kanakarajan, “A multimodal approach for extracting content descriptive metadata from 14 lecture videos,” J. Intell. Inf. Syst. , vol. 46, no. 1, p. 121–145, 2016. [Online]. Available: https://doi.org/10.1007/s10844-015-0356-5

    2016 doi

  20. [28]

    Automatic document metadata extraction based on deep networks,

    R. Liu, L. Gao, D. An, Z. Jiang, and Z. Tang, “Automatic document metadata extraction based on deep networks,” in Natural Language Processing and Chinese Computing , X. Huang, J. Jiang, D. Zhao, Y . Feng, and Y . Hong, Eds. Cham: Springer International Publishing, 2018, pp. 305–317

    2018

  21. [29]

    Arctic: metadata extraction from scientific papers in pdf using two-layer crf,

    A. Souza, V . Moreira, and C. Heuser, “Arctic: metadata extraction from scientific papers in pdf using two-layer crf,” in Proceedings of the 2014 ACM symposium on document engineering , 2014, pp. 121–130

    2014

  22. [31]

    Bi-lstm-crf sequence labeling for keyphrase extraction from scholarly documents,

    R. Alzaidy, C. Caragea, and C. L. Giles, “Bi-lstm-crf sequence labeling for keyphrase extraction from scholarly documents,” in The world wide web conference, 2019, pp. 2551–2557

    2019

  23. [32]

    Grantex- tractor: Accurate grant support information extraction from biomedical fulltext based on bi-lstm-crf,

    S. Dai, Y . Ding, Z. Zhang, W. Zuo, X. Huang, and S. Zhu, “Grantex- tractor: Accurate grant support information extraction from biomedical fulltext based on bi-lstm-crf,” IEEE/ACM transactions on computational biology and bioinformatics , vol. 18, no. 1, pp. 205–215, 2019

    2019

  24. [33]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 2961–2969

    2017

  25. [34]

    Aggregated residual transformations for deep neural networks,

    S. Xie, R. Girshick, P. Doll ´ar, Z. Tu, and K. He, “Aggregated residual transformations for deep neural networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1492– 1500

    2017

  26. [35]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2017, pp. 2117–2125

    2017

  27. [36]

    A survey on transfer learning,

    S. J. Pan and Q. Yang, “A survey on transfer learning,” IEEE Trans- actions on knowledge and data engineering , vol. 22, no. 10, pp. 1345– 1359, 2010

    2010

  28. [37]

    Detectron2,

    Y . Wu, A. Kirillov, F. Massa, W.-Y . Lo, and R. Girshick, “Detectron2,” https://github.com/facebookresearch/detectron2, 2019

    2019

  29. [38]

    Publaynet: largest dataset ever for document layout analysis,

    X. Zhong, J. Tang, and A. J. Yepes, “Publaynet: largest dataset ever for document layout analysis,” in 2019 International Conference on Document Analysis and Recognition (ICDAR) . IEEE, 2019, pp. 1015– 1022

    2019

  30. [39]

    Cermine: Automatic extraction of structured metadata from scientific literature,

    D. Tkaczyk, P. Szostek, M. Fedoryszak, P. J. Dendek, and L. Bolikowski, “Cermine: Automatic extraction of structured metadata from scientific literature,” Int. J. Doc. Anal. Recognit. , vol. 18, no. 4, p. 317–335,

  31. [40]

    Towards better ud parsing: Deep contextualized word embeddings, ensemble, and treebank concatenation,

    W. Che, Y . Liu, Y . Wang, B. Zheng, and T. Liu, “Towards better ud parsing: Deep contextualized word embeddings, ensemble, and treebank concatenation,” arXiv preprint arXiv:1807.03121 , 2018

    2018 arXiv

  32. [41]

    S2orc: The semantic scholar open research corpus,

    K. Lo, L. L. Wang, M. Neumann, R. Kinney, and D. S. Weld, “S2orc: The semantic scholar open research corpus,” arXiv preprint arXiv:1911.02782, 2019

    1911 arXiv

  33. [42]

    Crossref: The sus- tainable source of community-owned scholarly metadata,

    G. Hendricks, D. Tkaczyk, J. Lin, and P. Feeney, “Crossref: The sus- tainable source of community-owned scholarly metadata,” Quantitative Science Studies, vol. 1, no. 1, pp. 414–427, 2020

    2020

  34. [2015]

    Available: https://doi.org/10.1007/s10032-015-0249-8

    [Online]. Available: https://doi.org/10.1007/s10032-015-0249-8

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.