Pith. sign in

REVIEW 5 major objections 5 minor 48 references

A document processing pipeline for the construction of a dataset for topic modeling based on the judgments of the Italian Supreme Court

T0 review · 5 major / 5 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 A six-step computer-vision and NLP pipeline turns scanned Italian Supreme Court judgments into an anonymized corpus on which BERTopic finds up to 48 coherent, diverse legal topics, outperforming OCR-only processing.

desk verdict Solid OCR/DLA engineering and a real new Italian legal corpus, but the paper's central claim that the pipeline improves topic modeling is contradicted by its own Table 13 and confounded by an extra filtering step. read the letter →

arxiv 2505.08439 v1 pith:MP3MIPS3 submitted 2025-05-13 cs.CL

Matteo Marulli , Glauco Panattoni , Marco Bertini This is my paper · ORCID
classification cs.CL
keywords ItalianSupremeCourtjudgmentsdocumentlayoutanalysisopticalcharacterrecognitiontextanonymizationtopicmodelingBERTopicnamedentitylegalNLP
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 claims that a computer-vision document pipeline can turn scanned Italian Supreme Court judgments into a dataset that supports better topic modeling than the same texts run through OCR alone. The pipeline detects the layout of each page (titles, section headers, footers, paragraphs), extracts the text line by line, and anonymizes names, companies, locations, dates, and identifiers before any modeling. On this processed corpus, BERTopic finds up to 48 coherent, diverse topics, while on the OCR-only corpus it struggles past roughly a dozen. The authors also claim that anonymization is not just a privacy step but improves semantic quality, because named entities no longer dominate the terms that define each topic. If the claim holds, the paper contributes both a reproducible method for building shareable legal corpora and evidence that document structure should be treated as part of topic-modeling preprocessing rather than optional cleanup.

What carries the argument

The load-bearing mechanism is paragraph-level segmentation driven by document layout analysis. A YOLOv8x detector, fine-tuned from the DocLayNet annotation scheme, finds titles, section headers, footers, and body paragraphs, orders them in reading order, and crops each one; paragraph-sized units fit within BERT's 512-token context while preserving complete information, which whole-page text truncates and line-level text fragments. A second YOLOv8x locates text lines, TrOCR transcribes them, and GLiNER in an Italian PII-tuned variant replaces sensitive entities with tags such as ⟨PERSONA⟩ and ⟨LOCALITÀ⟩, tags that recur across documents, receive low c-TF-IDF weight, and therefore drop out of topic definitions. The evaluation machinery is BERTopic with UMAP and HDBSCAN clustering plus c-TF-IDF-bm25 topic representation, scored by the topic-diversity and Cv-coherence metrics across $K$ from 2 to 50.

What would settle it

Rerun the paper's comparison with the same paragraph-length filter applied to the OCR-only text: if topic diversity and coherence then match the pipeline condition, the claimed benefit of layout analysis is disproven. A second check is to have another legal expert independently label and summarize the same topics and recompute BERTScore against both references; if agreement with the second expert falls far below the reported figures, the LLM-quality claim is an artifact of the single reference.

Watch

Extended reading notes

Core claim

The paper's central claim is that document layout analysis, high-accuracy OCR, and entity anonymization, chained into one pipeline, produce a corpus on which unsupervised topic modeling is better than on the same documents extracted by OCR alone. The authors report DLA detection at mAP@50 of 0.964 and mAP@50-95 of 0.800, an OCR text detector at mAP@50-95 of 0.9022, and TrOCR recognition at a character error rate of 0.0047 and word error rate of 0.0248. On the full-pipeline dataset, BERTopic with an Italian legal-domain embedding model reaches $K=48$ topics at topic diversity 0.6803 and coherence 0.6809, while the OCR-only version saturates near $K=15$; the same pattern holds with a general-purpose Italian embedding model ($K=47$ versus $K=12$). Anonymization contributes to this: replacing names and other entities with recurring placeholders lowers their cluster-level term weights, so topics are defined by legal substance rather than by parties, companies, and locations. Finally, the paper claims that large language models can interpret the discovered topics, with Claude Sonnet 3.7 reaching BERTScore F1 of 0.8119 for labels and 0.9130 for summaries against a legal expert's reference.

Load-bearing premise

The central claim rests on assuming that the topic-modeling gains come from the layout-analysis and segmentation step, even though the pipeline condition also filters out titles, footers, headers, and all short paragraphs on its side only, and it assumes that a single legal expert's labels and summaries are a reliable reference for scoring the language models.

Editorial extensions

If this is right

  • Legal scholars and NLP researchers get a reproducible route to a shareable, GDPR-compliant Italian legal corpus: scanned judgments become anonymized, structured JSON without manual transcription.
  • Topic models on the pipeline output stay interpretable at much finer granularity (up to roughly 48 topics) than on OCR-only text, so recurring legal themes such as appeal grounds, computer fraud, and press defamation can be traced across the corpus.
  • The benefit transfers across at least two embedding strategies, so the improvement is tied to the document representation, not to one specific embedding model.
  • Anonymization does double duty: it protects personal data and, because entity placeholders recur across documents, it prevents names and companies from dominating topic keywords.
  • Large language models can draft topic labels and summaries close to an expert's reference (BERTScore F1 of about 0.81 for labels and 0.91 for summaries with Claude 3.7), making expert interpretation a verification step rather than a bottleneck.

Reading between the lines

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

  • A cheap version of the claimed effect may exist: the pipeline condition also removes titles, footers, headers, and all short paragraphs, so replicating that length filter on OCR-only text at near-zero cost would show whether layout analysis itself is what matters.
  • If the driver is paragraph length and self-containedness rather than layout semantics, the same recipe should transfer to other dense, long-form legal corpora in other languages, where the bottleneck is annotated layout data.
  • The corpus has an unexploited longitudinal structure: assigning each judgment its year would let the same pipeline measure doctrinal drift, such as the growth of computer-crime jurisprudence, which the paper does not attempt.
  • A second expert's annotations would likely shift the reported BERTScore figures, so those values should be read as agreement with one reference rather than an absolute quality ceiling for LLM topic interpretation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The manuscript describes an end-to-end pipeline that converts PDF judgments of the Italian Supreme Court into an anonymized, topic-modeling-ready dataset. The pipeline combines YOLOv8x-based document layout analysis, YOLOv8x line detection plus TrOCR recognition, and zero-shot GLiNER anonymization. The authors report strong module-level results (DLA mAP@50-95 = 0.800, OCR WER = 0.0248), then use BERTopic with legal-domain embeddings to compare topic diversity and coherence on pipeline-processed versus OCR-only text, and they evaluate LLM-generated topic labels and summaries against a legal expert using BERTScore. The central claim is that the full pipeline improves topic modeling over OCR-only processing.

Significance. If the comparative claim were valid, the paper would provide a valuable template for constructing privacy-compliant Italian legal topic-modeling datasets and a useful evaluation of LLM-based topic interpretation. The module-level evaluation is concrete, and the DLA and OCR numbers are strong; the authors also explicitly acknowledge the absence of a quantitative anonymization benchmark. However, the headline comparison is not supported by the paper's own results, as detailed below.

major comments (5)
  1. [Table 13, §5.3.3, and Abstract] For the primary embedding model (distil-ita-legal-bert), the OCR-only condition achieves higher topic diversity (0.7 vs 0.6803) and higher coherence (0.6878 vs 0.6809) than the pipeline condition. This directly contradicts the claim in §5.3.3 that 'the dataset generated through our segmentation pipeline yielded superior results' and the abstract's statement that the dataset surpasses 'traditional OCR-only approaches.' The abstract's reported values (0.6198 and 0.6638) do not appear in any row of Table 13, nor in the surrounding text, so the reader cannot determine which experiment produced the headline numbers.
  2. [§4.7 and §5.3.3] The pipeline-versus-OCR-only comparison is confounded: the pipeline condition removes titles, page footers, section headers, and 'short or uninformative paragraphs,' retaining only the right-hand tail of the paragraph-length distribution (Figure 15), whereas the OCR-only condition is said to receive 'no further refinement.' Any observed difference could therefore be caused by the length/content filter rather than by document layout analysis and segmentation. A valid test would hold the filtering step fixed across both conditions.
  3. [Table 13 and §5.3.3] The number of topics K is a free parameter swept from 2 to 50, yet Table 13 reports a single operating point per condition (K = 48, 15, 47, 12) without specifying the selection rule. Topic diversity and topic coherence are strongly K-dependent, so comparing results at different K values does not support the claim that the pipeline enables more topics at acceptable quality. The paper should report full K-sweep curves or compare at matched K with a pre-specified selection criterion.
  4. [§5.3.4] The anonymization component, which is part of the claimed contribution of a GDPR-compliant dataset, is not quantitatively evaluated: the authors state that the GLiNER model 'lacks a labeled dataset for quantitative performance evaluation' and rely on qualitative observations. Without precision/recall on a held-out set of Italian legal documents, the claims that the pipeline is privacy-compliant and that anonymization prevents entity-dominated topics are not established.
  5. [§5.4.1] The BERTScore evaluation uses a single legal expert's labels and summaries as the reference, with no inter-annotator agreement or second rater. The resulting F1 scores (e.g., 0.9130 for Claude) should therefore be interpreted as agreement with one expert rather than as general semantic quality; the paper overstates this as demonstrating 'comparable semantic quality.'
minor comments (5)
  1. [§2.4] There is an unresolved citation placeholder 'Licari et al. [ ? ]' in the text, and the reference list does not contain a corresponding entry.
  2. [Table 8] The table states that entities are grouped into 'five categories' but lists six entity types: Organization, Person, Location, Email, DATE, and ID.
  3. [§4 and §5] The section headings contain typos: 'Methedology' should be 'Methodology' and 'Evalutation' should be 'Evaluation.'
  4. [Conclusion] The conclusion repeats the unreported values 'topic diversity: 0.6198, topic coherence (Cv): 0.663'; these should be reconciled with Table 13 or the table should be completed with the corresponding experiment.
  5. [Table 13] Because UMAP and HDBSCAN are stochastic, single runs may not be stable; the paper should report seeds or repeated runs to support the reported topic modeling numbers.

Circularity Check

2 steps flagged · score 6.0 of 10

The central topic-modeling claim is a fitted, selected result: the abstract's 0.6198/0.6638 appear in no table row, and for the primary embedding model Table 13's OCR-only row beats the pipeline on both metrics.

  1. fitted input called prediction [Abstract; Section 5.3.3; Table 13; Conclusion]
    ""Compared to OCR-only methods, our dataset improved topic modeling with a diversity score of 0.6198 and a coherence score of 0.6638." ... "For each version, we conducted a series of experiments to assess topic modeling performance. In particular, we varied the number of topics K between 2 and 50, and computed two key metrics: topic diversity and topic coherence.""

    Table 13 is the only quantitative pipeline-versus-OCR comparison, and it reports selected K rows. For the primary embedding model, distil-ita-legal-bert, the no-DLA row has TD=0.7 and Cv=0.6878 at K=15, both higher than the pipeline row's TD=0.6803 and Cv=0.6809 at K=48. The abstract's claimed values 0.6198/0.6638 appear in no row. Because K was swept from 2 to 50 with no stated selection rule, the reported K values and headline metrics are post hoc selections, not predictions generated by the pipeline. The claimed improvement is therefore a fitted-value summary of chosen runs, not a derived consequence of the document processing pipeline.

  2. self definitional [Section 5.3.3; Section 4.7]
    ""The second was generated using our full pipeline, then filtered to remove nonessential content—such as titles, page footers, section headers, and short or uninformative paragraphs, which often contain run-on or fragmented text. The filtered dataset retains high-quality paragraphs, specifically those in the right-hand tail of the logarithmic distribution of tokenized paragraph lengths, ensuring a focus on content-rich text.""

    The condition labeled 'DLA segmentation' is defined to include, in addition to the DLA/OCR/NER pipeline, a post hoc filter that removes titles, headers, footers, and short paragraphs and keeps only the right-hand tail of paragraph lengths. The 'OCR-only' baseline is defined as text without this refinement. Any improvement in BERTopic coherence or diversity under the 'pipeline' condition is therefore attributable, by construction, to the length/content filter rather than to layout segmentation. The paper's conclusion that 'our segmentation strategy plays a critical role in improving the quality of topic modeling' restates an input choice embedded in the experimental condition instead of testing an independent effect of the pipeline.

full rationale

No load-bearing self-citation chain is present: the DLA and OCR metrics (mAP, CER, WER) are self-contained evaluations on the authors' own annotated test sets and are not circular. The circularity concerns the paper's central comparative claim. The abstract and conclusion assert that the full pipeline improves topic modeling over OCR-only text, but the only reported comparison, Table 13, contradicts this for the primary embedding model, and the abstract's numeric values do not appear in any table row. The K sweep from 2 to 50 with selected reporting turns the headline scores into fitted outputs rather than independent predictions. In addition, the 'pipeline' condition is not merely DLA segmentation: it also applies a right-tail paragraph-length filter and removes the very elements DLA detects, so the comparison cannot isolate the claimed mechanism. These two issues make the central contribution partially circular or fitted-value reporting. The single-expert LLM evaluation and BERTScore validity are separate methodological limitations, not circularity; they do not affect this score further.

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

The paper introduces no new theoretical entities. Its central claims rest on a set of fitted hyperparameters (K, paragraph filter, GLiNER threshold, BERTopic diversity) and on unvalidated domain assumptions (random sampling, high-quality scans, single-expert ground truth).

free parameters (5)
  • Paragraph length filtering cutoff (right-hand tail of tokenized length distribution) = not specified; quantile-based (first quartile, median, third quartile shown in Fig. 15)
    Applied only to the full-pipeline dataset before topic modeling (section 5.3.3); it is the main difference from the OCR-only condition and may drive the reported topic-modeling changes.
  • Number of topics K = 48 (distil-legal, with pipeline), 15 (distil-legal, without); 47 and 12 for the sentence-bert rows
    K was swept from 2 to 50 and the run inside the chosen diversity and coherence boundaries was reported (Table 13); the abstract's 0.6198 and 0.6638 values do not appear in any table row.
  • GLiNER entity confidence threshold = not reported
    Anonymization is described as thresholding-based (sections 5.3.4 and Conclusion), but the threshold value and its effect on false negatives are not given.
  • BERTopic diversity parameter for topic representation = 0.35
    Table 9 sets diversity=0.35, directly shaping the topic diversity metric (TD) that is a headline result.
  • UMAP and HDBSCAN hyperparameters = UMAP n_components=5, min_dist=0.0, n_neighbors=5; HDBSCAN min_cluster_size=5, min_samples=5
    Chosen without sensitivity analysis (Table 9); clustering output and hence topic metrics depend on them.
assumptions (5)
  • domain assumption The scanned judgments from Italgiure are of high quality and representative of the population of Supreme Court judgments.
    Stated in section 3.1; the claim that the 307 judgments were randomly chosen is not backed by a detailed sampling protocol.
  • domain assumption Paragraph-level segmentation and the 512-token BERT context limit motivate the DLA design.
    Section 3.2 argues paragraph-level boxes balance truncation and fragmentation; this choice is not empirically compared to line- or sentence-level alternatives.
  • domain assumption A single legal expert's topic labels and summaries are a valid ground truth for BERTScore evaluation.
    Section 5.4.1 uses one expert as reference; no inter-annotator agreement or second expert is reported.
  • ad hoc to paper Zero-shot GLiNER with Italian PII prompts produces anonymization sufficient for GDPR compliance.
    Sections 4.5 and 5.3.4 assert practical GDPR compliance without a quantitative evaluation; the authors acknowledge false negatives.
  • standard math The c-TF-IDF-bm25 and standard NPMI/Cv coherence formulas are applicable as implemented.
    Section 5.1.3 restates standard definitions from cited works; correctness of the implementation is not verified here.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A document processing pipeline for the construction of a dataset for topic modeling based on the judgments of the Italian Supreme Court." pith.science (2026). https://pith.science/paper/MP3MIPS3

@misc{pith2026250508439,
  author       = {Pith},
  title        = {Pith review of: A document processing pipeline for the construction of a dataset for topic modeling based on the judgments of the Italian Supreme Court},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MP3MIPS3}},
  note         = {Machine review of arXiv:2505.08439}
}
read the original abstract

Topic modeling in Italian legal research is hindered by the lack of public datasets, limiting the analysis of legal themes in Supreme Court judgments. To address this, we developed a document processing pipeline that produces an anonymized dataset optimized for topic modeling. The pipeline integrates document layout analysis (YOLOv8x), optical character recognition, and text anonymization. The DLA module achieved a mAP@50 of 0.964 and a mAP@50-95 of 0.800. The OCR detector reached a mAP@50-95 of 0.9022, and the text recognizer (TrOCR) obtained a character error rate of 0.0047 and a word error rate of 0.0248. Compared to OCR-only methods, our dataset improved topic modeling with a diversity score of 0.6198 and a coherence score of 0.6638. We applied BERTopic to extract topics and used large language models to generate labels and summaries. Outputs were evaluated against domain expert interpretations. Claude Sonnet 3.7 achieved a BERTScore F1 of 0.8119 for labeling and 0.9130 for summarization.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 33 canonical work pages

  1. [1]

    The elements of statistical learning: Data mining, inference, and prediction, 9–41 (2009)

    Hastie, T., Tibshirani, R., Friedman, J., Hastie, T., Tibshirani, R., Friedman, J.: Overview of supervised learning. The elements of statistical learning: Data mining, inference, and prediction, 9–41 (2009)

    work page 2009

  2. [2]

    In: 2018 Federated Conference on Computer Science and Information Systems (FedCSIS), pp

    Undavia, S., Meyers, A., Ortega, J.E.: A comparative study of classifying legal documents with neural networks. In: 2018 Federated Conference on Computer Science and Information Systems (FedCSIS), pp. 515–522 (2018)

    work page 2018

  3. [3]

    In: Legal Knowledge and Information Systems, pp

    Clavi´ e, B., Alphonsus, M.: The unreasonable effectiveness of the baseline: dis- cussing svms in legal text classification. In: Legal Knowledge and Information Systems, pp. 58–61. IOS Press, ??? (2021)

    work page 2021

  4. [4]

    Procedia Computer Science 13, 53–59 (2012)

    Thammaboosadee, S., Watanapa, B., Charoenkitkarn, N.: A framework of multi- stage classifier for identifying criminal law sentences. Procedia Computer Science 13, 53–59 (2012)

    work page 2012

  5. [5]

    Artificial Intelligence Review 53(2), 907–948 (2020)

    Solorio-Fern´ andez, S., Carrasco-Ochoa, J.A., Mart´ ınez-Trinidad, J.F.: A review of unsupervised feature selection methods. Artificial Intelligence Review 53(2), 907–948 (2020)

    work page 2020

  6. [6]

    In: Mexican International Conference on Artificial Intelligence, pp

    Prince-Tritto, P., Ponce, H.: Exploring the challenges and limitations of unsu- pervised machine learning approaches in legal concepts discovery. In: Mexican International Conference on Artificial Intelligence, pp. 52–67 (2023). Springer

    work page 2023

  7. [7]

    Information Systems 112, 102131 (2023) 40

    Abdelrazek, A., Eid, Y., Gawish, E., Medhat, W., Hassan, A.: Topic modeling algorithms and applications: A survey. Information Systems 112, 102131 (2023) 40

    work page 2023

  8. [8]

    A Practical Guide, 1st Ed., Cham: Springer International Publishing 10(3152676), 10–5555 (2017)

    Voigt, P., Bussche, A.: The eu general data protection regulation (gdpr). A Practical Guide, 1st Ed., Cham: Springer International Publishing 10(3152676), 10–5555 (2017)

    work page 2017

Show all 48 references
  1. [9]

    arXiv preprint arXiv:2011.13534 (2020)

    Subramani, N., Matton, A., Greaves, M., Lam, A.: A survey of deep learning approaches for ocr and document understanding. arXiv preprint arXiv:2011.13534 (2020)

    2020 arXiv

  2. [10]

    ACM Computing Surveys (CSUR) 52(6), 1–36 (2019)

    Binmakhashen, G.M., Mahmoud, S.A.: Document layout analysis: a comprehen- sive survey. ACM Computing Surveys (CSUR) 52(6), 1–36 (2019)

    2019

  3. [11]

    arXiv preprint arXiv:2303.18223 (2023)

    Zhao, W.X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., et al.: A survey of large language models. arXiv preprint arXiv:2303.18223 (2023)

    2023 arXiv

  4. [12]

    In: Proceedings of the 10th Italian Conference on Computational Linguistics (CLiC-it 2024), pp

    Giaconia, A., Chiariello, V., Passarotti, M.: Topic modeling for auditing pur- poses in the banking sector. In: Proceedings of the 10th Italian Conference on Computational Linguistics (CLiC-it 2024), pp. 1030–1035 (2024)

    2024

  5. [13]

    In: Data Science and Social Research II: Methods, Technologies and Applications, pp

    Cataldo, R., Grassia, M.G., Marino, M., Mazza, R., Pastena, V., Zavarrone, E.: Divorce in italy: a textual analysis of cassation judgment. In: Data Science and Social Research II: Methods, Technologies and Applications, pp. 269–280 (2021). Springer

    2021

  6. [14]

    Artificial Intelligence and Law 32(4), 1045–1074 (2024)

    Vianna, D., Moura, E.S., Silva, A.S.: A topic discovery approach for unsupervised organization of legal document collections. Artificial Intelligence and Law 32(4), 1045–1074 (2024)

    2024

  7. [15]

    Silveira, R., Fernandes, C.G., Araujo Monteiro Neto, J., Furtado, V., Pimentel Filho, J.E.: Topic modelling of legal documents via legal-bert (2021)

    2021

  8. [16]

    arXiv e-prints, 2401 (2023)

    Rang, M., Bi, Z., Liu, C., Wang, Y., Han, K.: An empirical study of scaling law for ocr. arXiv e-prints, 2401 (2023)

    2023

  9. [17]

    In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pp

    Audebert, N., Herold, C., Slimani, K., Vidal, C.: Multimodal deep networks for text and image-based document classification. In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pp. 427–443 (2019). Springer

    2019

  10. [18]

    Computer Systems Science & Engineering 48(3) (2024)

    Baimakhanova, A., Zhumadillayeva, A., Mukhametzhanova, B., Glazyrina, N., Niyazova, R., Zhunissov, N., Sambetbayeva, A.: Multimodal deep neural networks for digitized document classification. Computer Systems Science & Engineering 48(3) (2024)

    2024

  11. [19]

    JAMIA open 5(2), 045 (2022) 41

    Hsu, E., Malagaris, I., Kuo, Y.-F., Sultana, R., Roberts, K.: Deep learning-based nlp data pipeline for ehr-scanned document information extraction. JAMIA open 5(2), 045 (2022) 41

    2022

  12. [20]

    arXiv preprint arXiv:2406.15032 (2024)

    Salierno, G., Bert` e, R., Attias, L., Morrone, C., Pettazzoni, D., Battisti, D.: Gius- berto: A legal language model for personal data de-identification in italian court of auditors decisions. arXiv preprint arXiv:2406.15032 (2024)

    2024 arXiv

  13. [21]

    Zaratiana, U., Tomeh, N., Holat, P., Charnois, T.: GLiNER: Generalist Model for Named Entity Recognition using Bidirectional Transformer (2023)

    2023

  14. [22]

    doi: 10.48550

    Pham, C., Hoyle, A., Sun, S., Resnik, P., Iyyer, M.: Topicgpt: A prompt-based topic modeling framework. doi: 10.48550. arXiv preprint arXiv.2311.01449 (2024)

    2024 arXiv

  15. [23]

    In: The 20th World Congress of the International Fuzzy Systems Association (2023)

    Rijcken, E., Scheepers, F., Zervanou, K., Spruit, M., Mosteiro, P., Kaymak, U.: Towards interpreting topic models with chatgpt. In: The 20th World Congress of the International Fuzzy Systems Association (2023)

    2023

  16. [24]

    arXiv preprint arXiv:1609.08144 (2016)

    Wu, Y., Schuster, M., Chen, Z., Le, Q.V., Norouzi, M., Macherey, W., Krikun, M., Cao, Y., Gao, Q., Macherey, K., et al.: Google’s neural machine translation system: Bridging the gap between human and machine translation. arXiv preprint arXiv:1609.08144 (2016)

    2016 arXiv

  17. [25]

    https://labelstud.io

    Studio, L.: Label Studio – Open Source Data Labeling Tool. https://labelstud.io. Ultimo accesso: 18 febbraio 2025

    2025

  18. [26]

    In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp

    Pfitzmann, B., Auer, C., Dolfi, M., Nassar, A.S., Staar, P.: Doclaynet: A large human-annotated dataset for document-layout segmentation. In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 3743–3751 (2022)

    2022

  19. [27]

    In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2016)

    Redmon, J.: You only look once: Unified, real-time object detection. In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2016)

    2016

  20. [28]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Li, M., Lv, T., Chen, J., Cui, L., Lu, Y., Florencio, D., Zhang, C., Li, Z., Wei, F.: Trocr: Transformer-based optical character recognition with pre-trained models. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, pp. 13094–13102 (2023)

    2023

  21. [29]

    arXiv preprint arXiv:2106.08254 (2021)

    Bao, H., Dong, L., Piao, S., Wei, F.: Beit: Bert pre-training of image transformers. arXiv preprint arXiv:2106.08254 (2021)

    2021 arXiv

  22. [30]

    arXiv preprint arXiv:1907.11692 364 (2019)

    Liu, Y.: Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692 364 (2019)

    2019 arXiv

  23. [31]

    https://huggingface.co/DeepMount00/ GLiNER PII ITA

    DeepMount00: GLiNER PII ITA. https://huggingface.co/DeepMount00/ GLiNER PII ITA. Ultimo accesso: 18 febbraio 2025

    2025

  24. [32]

    : pandas: a foundational python library for data analysis and statistics

    McKinney, W., et al. : pandas: a foundational python library for data analysis and statistics. Python for high performance and scientific computing 14(9), 1–9 42 (2011)

    2011

  25. [33]

    arXiv preprint arXiv:2203.05794 (2022)

    Grootendorst, M.: Bertopic: Neural topic modeling with a class-based tf-idf procedure. arXiv preprint arXiv:2203.05794 (2022)

    2022 arXiv

  26. [34]

    Computer Law & Security Review 52, 105908 (2024)

    Licari, D., Comand` e, G.: Italian-legal-bert models for improving natural language processing tasks in the italian legal domain. Computer Law & Security Review 52, 105908 (2024)

    2024

  27. [35]

    arXiv preprint arXiv:2109.02508 (2021)

    Ghojogh, B., Ghodsi, A., Karray, F., Crowley, M.: Uniform manifold approxima- tion and projection (umap) and its variants: tutorial and survey. arXiv preprint arXiv:2109.02508 (2021)

    2021 arXiv

  28. [36]

    : hdbscan: Hierarchical density based clustering

    McInnes, L., Healy, J., Astels, S., et al. : hdbscan: Hierarchical density based clustering. J. Open Source Softw. 2(11), 205 (2017)

    2017

  29. [37]

    In: 2019 International Engineering Conference (IEC), pp

    Qader, W.A., Ameen, M.M., Ahmed, B.I.: An overview of bag of words; importance, implementation, applications, and challenges. In: 2019 International Engineering Conference (IEC), pp. 200–204 (2019). IEEE

    2019

  30. [38]

    arXiv preprint arXiv:2407.21783 (2024)

    Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Yang, A., Fan, A., et al.: The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    2024 arXiv

  31. [39]

    https://www.anthropic.com/news/ claude-3-family

    Anthropic: Claude 3 Family Announcement. https://www.anthropic.com/news/ claude-3-family. Ultimo accesso: 18 febbraio 2025

    2025

  32. [40]

    arXiv preprint arXiv:2303.08774 (2023)

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F.L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al.: Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    2023 arXiv

  33. [41]

    In: 2020 International Conference on Systems, Sig- nals and Image Processing (IWSSIP), pp

    Padilla, R., Netto, S.L., Silva, E.A.B.: A survey on performance metrics for object-detection algorithms. In: 2020 International Conference on Systems, Sig- nals and Image Processing (IWSSIP), pp. 237–242 (2020). https://doi.org/10. 1109/IWSSIP48289.2020.9145130

    2020

  34. [42]

    In: Proceedings of the 6th International Workshop on Historical Document Imaging and Processing, pp

    Neudecker, C., Baierer, K., Gerber, M., Clausner, C., Antonacopoulos, A., Pletschacher, S.: A survey of ocr evaluation tools and metrics. In: Proceedings of the 6th International Workshop on Historical Document Imaging and Processing, pp. 13–18 (2021)

    2021

  35. [43]

    In: Proceedings of the Eighth ACM International Conference on Web Search and Data Mining, pp

    R¨ oder, M., Both, A., Hinneburg, A.: Exploring the space of topic coherence mea- sures. In: Proceedings of the Eighth ACM International Conference on Web Search and Data Mining, pp. 399–408 (2015)

    2015

  36. [44]

    Transactions of the Association for Computational Linguistics 8, 439–453 (2020) 43 https://doi.org/10.1162/tacl a 00325

    Dieng, A.B., Ruiz, F.J.R., Blei, D.M.: Topic modeling in embedding spaces. Transactions of the Association for Computational Linguistics 8, 439–453 (2020) 43 https://doi.org/10.1162/tacl a 00325

    2020 doi

  37. [45]

    arXiv preprint arXiv:2106.15971 (2021)

    Bilal, I.M., Wang, B., Liakata, M., Procter, R., Tsakalidis, A.: Evaluation of thematic coherence in microblogs. arXiv preprint arXiv:2106.15971 (2021)

    2021 arXiv

  38. [46]

    In: Webber, B., Cohn, T., He, Y., Liu, Y

    Wu, X., Li, C., Zhu, Y., Miao, Y.: Short text topic modeling with topic dis- tribution quantization and negative sampling decoder. In: Webber, B., Cohn, T., He, Y., Liu, Y. (eds.) Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), p...

    2020 doi

  39. [47]

    arXiv preprint arXiv:1904.09675 (2019)

    Zhang, T., Kishore, V., Wu, F., Weinberger, K.Q., Artzi, Y.: Bertscore: Evaluat- ing text generation with bert. arXiv preprint arXiv:1904.09675 (2019)

    2019 arXiv

  40. [48]

    https://huggingface.co/ nickprock/sentence-bert-base-italian-uncased

    nickprock: sentence-bert-base-italian-uncased. https://huggingface.co/ nickprock/sentence-bert-base-italian-uncased. Ultimo accesso: 18 febbraio 2025 44 T able 15 Comparison of model-generated labels and summaries for topic 0 Source Label Summary Expert Ricorso in Cassazione p...

    2025

Pith tools

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