Pith. sign in

REVIEW 4 major objections 4 minor 63 references

PBa-LLM: Privacy- and Bias-aware NLP using Named-Entity Recognition (NER)

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that NER-based removal of person and location entities from resume biographies preserves—and sometimes improves—occupancy prediction and scoring accuracy, while adding a bias-reduction step balances the top-100 gender…

desk verdict A useful empirical comparison of NER anonymizers for resume scoring, but the privacy claims outrun the measurements. read the letter →

arxiv 2507.02966 v2 pith:TA5GXZTH submitted 2025-06-30 cs.CL cs.AIcs.CRcs.LG

Gonzalo Mancera , Aythami Morales , Julian Fierrez , Ruben Tolosana , Alejandro Penna , Miguel Lopez-Duran , Francisco Jurado , Alvaro Ortigosa This is my paper
classification cs.CLcs.AIcs.CRcs.LG
keywords privacybiasfairnessnamedentityrecognitionlargelanguagemodelsanonymizationautomatedrecruitmentresponsibleAI
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 is trying to establish that named-entity recognition can serve as a privacy-preserving preprocessing step for language models without sacrificing accuracy, and that combining it with bias-aware training yields models that are both more private and more gender-fair. It demonstrates this in automated recruitment using 24,000 synthetic resumes, two text encoders, and six anonymization tools. The central claim is that removing person and location entities changes little or improves task performance, and that adding a debiasing technique moves gender balance in top-100 candidate lists from about 67/33 to about 50/50. A sympathetic reader would care because it suggests that privacy and fairness interventions can be added to high-stakes NLP pipelines at low performance cost, which bears directly on legal and ethical constraints such as the EU AI Act.

What carries the argument

The central machinery is a Named-Entity Recognition (NER) anonymization module that detects and masks sensitive tokens in resume biographies before training. Six different NER engines are compared—Presidio, FLAIR, a BERT model fine-tuned on CoNLL-2003, GPT-3.5, GPT-4, and GPT-4o-mini—and each one removes person (PER) and location (LOC) entities. The downstream model freezes a Transformer encoder's weights, averages its token embeddings into a text feature vector, fuses that vector with the candidate's structured competency attributes, and then predicts both occupational sector and an expert-style score.

What would settle it

Manually scan a random sample of the anonymized FairCVdb biographies for surviving person and location names, and compute the miss rate per anonymizer; if the miss rate is nontrivial, the tools have not fully removed the sensitive information regardless of downstream accuracy. A complementary test is a re-identification or membership-inference attempt on the masked texts.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that masking person and location entities detected by NER preserves, and sometimes slightly improves, the performance of occupation-prediction and candidate-scoring models built on BERT and RoBERTa, while adding a bias-reduction step produces PBa-LLMs whose top-100 gender split becomes approximately balanced even when trained on gender-biased labels. The paper states that in the worst case performance remains unchanged, while in the best case privacy increases without compromising accuracy.

Load-bearing premise

The privacy guarantee rests on the assumption that the six NER modules catch nearly every name and location in resume-style biographies, since the paper reports benchmark scores on CoNLL-2003 but never measures what fraction of entities is actually missed on its 24,000-resume testbed.

Editorial extensions

If this is right

  • If correct, NER-based masking is a viable low-cost privacy layer for LLM pipelines in high-stakes domains, since it does not force a privacy/accuracy tradeoff.
  • Because the results hold across two encoders and six anonymizers, the performance-preservation effect is not tied to a single NER tool; even imperfect NER models appear sufficient on the utility side.
  • Adding the bias-aware step to anonymized text changes the gender composition of top-100 shortlists from roughly 67/33 to roughly 50/50 without hurting occupancy prediction, so privacy and bias mitigation can be combined in one pipeline.
  • The authors state that the framework is generally applicable beyond recruitment, making the two-stage recipe—entity masking plus representation-level debiasing—a candidate template for other LLM-based text domains.

Reading between the lines

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

  • A direct test the paper leaves implicit is to compute recall and precision of each NER module on the FairCVdb biographies themselves, since CoNLL-2003 benchmark scores do not guarantee in-domain coverage; if recall is low, the confidentiality claim is only as strong as the entities that were missed.
  • Even if the text is fully anonymized, the structured attribute vector still includes gender and ethnicity, so the privacy-aware models continue to receive demographic signals through a non-textual channel; a stricter privacy reading would require masking or dropping those attributes too.
  • The gender-balance improvement may partly follow from removing gendered language through NER plus the debiasing step; a useful extension would measure how much of the 67/33 to 50/50 shift is caused by each component alone.
  • The performance-neutrality result could be domain-sensitive: in healthcare or legal texts, news-trained NER models may miss entity types that are rare in CoNLL-2003, so the same experiment should be repeated where entity distributions differ.
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

4 major / 4 minor

Summary. The paper proposes a framework for privacy-preserving NLP based on named-entity recognition (NER). In the recruitment use case, short resume biographies from the FairCVdb dataset are anonymized by masking PER and LOC entities using six NER tools (Presidio, FLAIR, BERT-CoNLL2003, GPT-3.5, GPT-4, GPT-4o-mini). Two Transformer backbones (BERT and RoBERTa) are then trained on the original and anonymized biographies for two tasks: four-way occupancy prediction and scalar candidate scoring. The paper reports that NER-based anonymization maintains or slightly improves occupancy prediction accuracy, and that combining anonymization with the bias-aware training objective of Peña et al. [40] yields PBa-LLMs that keep gender proportions in top-100 candidate shortlists close to 50/50. The central claims are that NER anonymization preserves downstream task performance while safeguarding candidate confidentiality, and that PBa-LLMs do not degrade scoring performance while reducing gender bias.

Significance. If the performance-preservation result is reliable, the paper would provide a useful empirical comparison of six NER anonymizers in a high-stakes recruitment setting, covering two transformer architectures and two tasks. The use of an established multimodal benchmark (FairCVdb) and the breadth of anonymizers are strengths, and the paper is clear about the pipeline it evaluates. However, the privacy claim is not directly tested: the authors never measure in-domain NER recall/precision, they mask only PER and LOC while their own definition of sensitive information includes companies, and the structured attribute vector still carries gender and ethnicity into the model. The bias results are also largely inherited from the bias-aware objective of [40] by construction, and the scoring task is evaluated only through top-100 gender proportions without any scoring error metric. The empirical core is useful but the paper currently overstates its privacy guarantees.

major comments (4)
  1. [§3.1, §3.2, §4.2] The privacy claim is not supported by the reported experiments. Section 4.2 reports entity counts and CoNLL-2003 F1 scores above 80%, but neither measure establishes that the anonymized biographies are actually free of personally identifiable information. The paper never measures in-domain recall or precision on the FairCVdb biographies, so missed PER and LOC entities could remain in the text. Moreover, Section 3.1 defines sensitive information as including names, locations, and companies, while Section 3.2 and Section 4.2 state that the anonymizer masks only PER and LOC entities; therefore the formal condition z_i not in t_i^- is not achieved for company names, and the conclusion that the framework 'safeguards candidate confidentiality' goes beyond what is measured.
  2. [§4.1, Fig. 2] Even perfect NER removal from the biography would not remove demographic attributes from the model input, because the candidate competencies vector v_i includes gender and ethnicity (Section 4.1) and is fused with the text embedding in the scoring model (Fig. 2). The claim that anonymization 'increases privacy' is therefore only about one modality and should be stated as such; as written, the abstract and Section 5.1 imply a confidentiality guarantee that the architecture does not provide.
  3. [§5.2, Table 3] The paper claims that PBa-LLMs 'do not degrade the performance of occupation prediction models' and that the combined approach is useful for candidate scoring, but Table 3 reports only gender proportions in the top-100 shortlist. No RMSE, MAE, or equivalent error metric is given for the scoring task, and Table 2 covers only occupancy prediction. The scoring-performance claim is therefore unsupported and needs a direct evaluation on the scoring output y_hat.
  4. [§5.1, Tables 1-2] The occupancy results are computed on a single fixed 80/20 split with no confidence intervals, significance tests, or multiple runs. Most anonymized-versus-baseline differences are a few accuracy points, so the statement that 'in the worst case, performance remains unchanged' is not established; the observed fluctuations could be within run-to-run or split-to-split variability. Reporting mean and standard deviation over several seeds or splits is needed to support the performance-preservation claim.
minor comments (4)
  1. [Table 2] The caption says 'Gender (all rows except None (Baseline))' but the table has no gender column; the caption appears to be a leftover from an earlier draft and should be corrected.
  2. [Fig. 2] The sample biography contains the typo '20014-2017' for the employment period.
  3. [§4.2] The paper reports that all six NER modules exceed 80% on CoNLL-2003 but does not give the per-module F1 values or their standard deviations; a small table would make the anonymizer comparison more informative.
  4. [Reference list] Reference [14] contains a typo ('Europen Union') and the citation formatting for some entries (e.g., [31] as a bare arXiv ID) is inconsistent with the rest of the bibliography.

Circularity Check

2 steps flagged · score 4.0 of 10

Fairness result is inherited from the authors' own debiasing objective; privacy gain is definitional, but the core performance benchmark is independent.

  1. self definitional [Sec. 3.1, Sec. 4.2, Sec. 5.1]
    "Assume that there is sensitive information zi within ti that we want to protect. We use the anonymization function N to generate t−i, so zi ∉ t−i. ... The number of entities removed varies from 28,000 to 66,000 ... This information represents approximately between 4% and 9% of the total words ... In the worst case, performance remains unchanged, while in the best case, privacy is increased without compromising accuracy."

    By the paper's own formalization, privacy is defined as the removal of sensitive entities: the stated goal is the condition z_i ∉ t−_i, and the only evidence adduced about privacy is the count of removed entities. Therefore 'privacy is increased' is a restatement of the intervention itself, not an empirical measurement of confidentiality. No in-domain recall/precision on FairCVdb, no re-identification test, and no independent privacy metric are reported. Moreover, Sec. 3.1 defines z_i as including 'companies', while Fig. 2 and Sec. 4.2 state that only PER and LOC are masked, so even the paper's own formal condition z_i ∉ t−_i is not satisfied for ORG entities. The phrase 'safeguarding candidate confidentiality' goes beyond what the definitional entity-removal counts can support.

  2. self citation load bearing [Sec. 5.2, Table 3]
    "To address this, we will employ the bias-aware learning framework introduced in [39] and further detailed in [40], which represents a method to systematically remove gender characteristics from the model's internal representations. ... None (Bias-aware) 47.80% 52.20%"

    The gender-balance result of PBa-LLM is produced by the debiasing objective of [40], which by design 'systematically remove[s] gender characteristics'. Table 3 makes this explicit: the row 'None (Bias-aware)' already reaches 47.80/52.20 male/female without any NER anonymization, and every NER-based row lands at approximately the same 50/50 split. Thus the fairness outcome is not a discovery of the proposed NER-plus-debiasing combination; it is a construction-level consequence of applying the same authors' earlier method. Since [39,40] are prior works with overlapping authorship and the paper presents this inherited outcome as the fairness component of PBa-LLM, the load-bearing support for the fairness claim is a self-citation chain rather than an independently derived prediction.

full rationale

The core performance result of the paper is genuinely empirical and self-contained: it compares six external NER tools (Presidio, FLAIR, BERT-CoNLL2003, GPT-3.5, GPT-4, GPT-4o-mini) against a non-anonymized baseline on 24,000 resume biographies, and the occupancy/scoring accuracy tables stand as independent evidence that masking PER and LOC entities does not degrade the downstream tasks. That part is not circular. The circularity is concentrated in two softer claims. First, the conclusion that 'privacy is increased' is a definitional restatement of the entity-removal intervention; the paper never measures whether the anonymized biographies are actually free of re-identifying information or whether the NER modules perform well in the resume domain. Second, the PBa-LLM fairness result is inherited from the bias-aware objective of [40], which is by the same research group and is explicitly designed to remove gender characteristics; Table 3's 'None (Bias-aware)' row shows the same 50/50 gender split without any NER anonymization, so the fairness contribution reduces to the imported objective. Because the accuracy claim is independently grounded in external tools and a public benchmark, the overall circularity is moderate rather than total. Score 4 reflects that the central performance claim has independent content while the privacy and fairness components are partly self-definitional or self-citation-loaded.

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

The central empirical benchmark uses public data and external tools, but the PBa-LLM fairness results depend on a hand-tuned debiasing objective from the authors' companion paper, and the privacy claim depends on unverified NER recall in the resume domain.

free parameters (3)
  • Bias-reduction strength parameter from Peña et al. [40] = not reported in this paper
    The Table 3 gender balances are produced by the bias-aware objective defined in [40]; its trade-off weight is not given here, so the main fairness result inherits a tuned quantity from a companion paper.
  • Choice of entity types PER and LOC = PER, LOC only
    The anonymization scope is hand-chosen; companies and other sensitive entities mentioned in Fig 2 are not masked, and the paper does not justify the scope.
  • Top-100 shortlist threshold = 100
    The fairness metric in Table 3 depends on the shortlist size; the choice of 100 is arbitrary and not varied.
assumptions (3)
  • domain assumption Frozen pretrained transformer features are sufficient for the downstream tasks
    Sec 4.3 freezes BERT/RoBERTa and trains only the fusion MLP; the method's viability rests on this transfer assumption.
  • ad hoc to paper CoNLL-2003 metrics transfer to resume biographies
    Sec 4.2 asserts each NER module exceeds 80% on CoNLL-2003; no resume-domain recall or precision is reported, yet the privacy claim depends on high recall.
  • domain assumption The bias-aware method of [40] is valid and correctly implemented
    Sec 5.2 delegates the debiasing objective to prior work by the same group; the gender-balance results inherit this assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PBa-LLM: Privacy- and Bias-aware NLP using Named-Entity Recognition (NER)." pith.science (2026). https://pith.science/paper/TA5GXZTH

@misc{pith2026250702966,
  author       = {Pith},
  title        = {Pith review of: PBa-LLM: Privacy- and Bias-aware NLP using Named-Entity Recognition (NER)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TA5GXZTH}},
  note         = {Machine review of arXiv:2507.02966}
}
read the original abstract

The use of Natural Language Processing (NLP) in highstakes AI-based applications has increased significantly in recent years, especially since the emergence of Large Language Models (LLMs). However, despite their strong performance, LLMs introduce important legal/ ethical concerns, particularly regarding privacy, data protection, and transparency. Due to these concerns, this work explores the use of Named- Entity Recognition (NER) to facilitate the privacy-preserving training (or adaptation) of LLMs. We propose a framework that uses NER technologies to anonymize sensitive information in text data, such as personal identities or geographic locations. An evaluation of the proposed privacy-preserving learning framework was conducted to measure its impact on user privacy and system performance in a particular high-stakes and sensitive setup: AI-based resume scoring for recruitment processes. The study involved two language models (BERT and RoBERTa) and six anonymization algorithms (based on Presidio, FLAIR, BERT, and different versions of GPT) applied to a database of 24,000 candidate profiles. The findings indicate that the proposed privacy preservation techniques effectively maintain system performance while playing a critical role in safeguarding candidate confidentiality, thus promoting trust in the experimented scenario. On top of the proposed privacy-preserving approach, we also experiment applying an existing approach that reduces the gender bias in LLMs, thus finally obtaining our proposed Privacyand Bias-aware LLMs (PBa-LLMs). Note that the proposed PBa-LLMs have been evaluated in a particular setup (resume scoring), but are generally applicable to any other LLM-based AI application.

Figures

Figures reproduced from arXiv: 2507.02966 by the authors.

Figure 1
Figure 1. Graphical abstract of the framework evaluated in this work. 1 Introduction During the past decade, automated decision-making systems have increasingly permeated sensitive domains such as education [8, 6], healthcare [27, 53], finance [52], and recruitment [4, 43, 44], driven by rapid advances in AI technologies. As these systems assume tasks traditionally requiring human judgment, they raise critical questions about… view at source ↗
Figure 2
Figure 2. Learning framework for the automatic recruitment tool. The system takes a multimodal input resume, consisting of a candidate competencies vector v and a text biography. An anonymization module N removes two types of identities: locations (LOC) and persons (PER). A pre-trained Transformer processes the anonymized bi￾ographies to extract a feature vector f. This feature vector is fused with the candidate competencies … view at source ↗
Figure 2
Figure 2. This system is trained on the dataset D to accurately estimate a new score yˆi that approximates a target score yi (e.g., supervised labels produced by human resources experts) utilizing the input resume {vi , ti}. It is important to note that the scoring mechanism comprises: i) a pre-trained Transformer (e.g., BERT and RoBERTa in our experiments), which accepts a tokenized biography as input and generates a text em… view at source ↗
Figures from the paper (2 more)
Figure 3
Figure 3. Figure 3: Accuracy over epochs for the BERT-based occupancy prediction model. The Figure compares different implementations, including six anonymization modules over different two entities (PER and LOC). The baseline model, without anonymization module, is highlighted with a dot…
Figure 4
Figure 4. Figure 4: Accuracy over epochs for the RoBERTa-based occupancy prediction model. The figure compares different implementations, including six anonymization modules over two different entities (PER and LOC). The baseline model, without anonymization module, is highlighted with a …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 56 canonical work pages

  1. [40]

    arXiv:2506.11880 (2025)

    Peña, A., Fierrez, J., Morales, A., Mancera, G., Lopez, M., Tolosana, R.: Ad- dressing bias in LLMs: Strategies and application to fair AI-based recruitment. arXiv:2506.11880 (2025)

    arXiv 2025

  2. [1]

    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:2303.08774 (2023)

    arXiv 2023

  3. [2]

    Science347(6221), 509–514 (2015)

    Acquisti, A., Brandimarte, L., Loewenstein, G.: Privacy and human behavior in the age of information. Science347(6221), 509–514 (2015)

    work page 2015

  4. [3]

    In: Proceedings of the Confer- ence of the North American Chapter of the Association for Computational Lin- guistics (Demonstrations)

    Akbik, A., Bergmann, T., Blythe, D., Rasul, K., Schweter, S., Vollgraf, R.: FLAIR: An easy-to-use framework for state-of-the-art NLP. In: Proceedings of the Confer- ence of the North American Chapter of the Association for Computational Lin- guistics (Demonstrations). pp. 54–59 (2019)

    work page 2019

  5. [4]

    In: 2021 International Conference on Decision Aid Sciences and Application (DASA)

    Al-Alawi, A.I., Naureen, M., AlAlawi, E.I., Al-Hadad, A.A.N.: The role of arti- ficial intelligence in recruitment process decision-making. In: 2021 International Conference on Decision Aid Sciences and Application (DASA). pp. 197–203. IEEE (2021)

    work page 2021

  6. [5]

    Babych, B., Hartley, A.: Improving machine translation quality with automatic namedentityrecognition.In:Proceedingsofthe7thInternationalEAMTworkshop onMTandotherlanguagetechnologytools,ImprovingMTthroughotherlanguage technology tools, Resource and tools for building MT at EACL 2003 (2003)

    work page 2003

  7. [6]

    AI-based Multimodal Biometrics for Detecting Smartphone Distractions: Application to Online Learning

    Becerra, A., Daza, R., Cobos, R., Morales, A., Cukurova, M., Fierrez, J.: AI-based multimodal biometrics for detecting smartphone distractions: Application to online learning. arXiv:2506.17364 (2025)

    work page Pith review arXiv 2025

  8. [7]

    Journal of Artificial Intelligence Research71, 1137–1181 (2021)

    Cheng, L., Varshney, K.R., Liu, H.: Socially responsible AI algorithms: Issues, purposes, and challenges. Journal of Artificial Intelligence Research71, 1137–1181 (2021)

    work page 2021

Show all 63 references
  1. [8]

    In: AAAI Conf

    Daza, R., Morales, A., Tolosana, R., Gomez, L.F., Fierrez, J., Ortega-Garcia, J.: edBB-Demo: Biometrics and behavior analysis for online educational platforms. In: AAAI Conf. on Artificial Intelligence (AAAI). vol. 37, pp. 16422–1642 (February 2023)

    2023

  2. [9]

    In: Proceedings of the Conference on Fairness, Accountability, and Transparency

    De-Arteaga, M., Romanov, A., Wallach, H., Chayes, J., Borgs, C., Chouldechova, A., Geyik, S., Kenthapadi, K., Kalai, A.T.: Bias in bios: A case study of semantic representation bias in a high-stakes setting. In: Proceedings of the Conference on Fairness, Accountability, and Tr...

    2019

  3. [10]

    arXiv:2402.09225 (2024)

    DeAlcala, D., Morales, A., Mancera, G., Fierrez, J., Tolosana, R., Ortega-Garcia, J.: Is my data in your AI model? membership inference test with application to face images. arXiv:2402.09225 (2024)

    2024

  4. [11]

    In: IEEE Conf

    DeAlcala, D., Serna, I., Morales, A., Fierrez, J., Ortega-Garcia, J.: Measuring bias in AI models: An statistical approach introducing n-sigma. In: IEEE Conf. on Computers, Software, and Applications (COMPSAC). pp. 1167–1172 (June 2023)

    2023

  5. [12]

    Digital Journalism3(3), 398–415 (2015) 16 Mancera et al

    Diakopoulos, N.: Algorithmic accountability: Journalistic investigation of compu- tational power structures. Digital Journalism3(3), 398–415 (2015) 16 Mancera et al

    2015

  6. [13]

    In: Proceedings of the 3rd International Workshop on Emotion Awareness in Software Engineering

    Ding, J., Sun, H., Wang, X., Liu, X.: Entity-level sentiment analysis of issue com- ments. In: Proceedings of the 3rd International Workshop on Emotion Awareness in Software Engineering. pp. 7–13 (2018)

    2018

  7. [14]

    European Parliament: European Parlia- mentary Research Service2024/1689 (2024)

    Europen Union: Artificial intelligence act. European Parliament: European Parlia- mentary Research Service2024/1689 (2024)

    2024

  8. [15]

    Journal of Web Semantics9(4), 434–452 (2011)

    Fernández, M., Cantador, I., López, V., Vallet, D., Castells, P., Motta, E.: Se- mantically enhanced information retrieval: An ontology-based approach. Journal of Web Semantics9(4), 434–452 (2011)

    2011

  9. [16]

    In: Proceedings of the 59th Annual Meeting of the ACL

    Fu, J., Huang, X.J., Liu, P.: Spanner: Named entity re-/recognition as span pre- diction. In: Proceedings of the 59th Annual Meeting of the ACL. pp. 7183–7195 (2021)

    2021

  10. [17]

    International Journal of Mechanical Engineering and Technology9(7), 63–70 (2018)

    Geetha, R., Bhanu, S.R.D.: Recruitment through artificial intelligence: a concep- tual study. International Journal of Mechanical Engineering and Technology9(7), 63–70 (2018)

    2018

  11. [18]

    In: IEEE Conf

    Ghafourian, M., Fierrez, J., Gomez, L.F., Vera-Rodriguez, R., Morales, A., Rezgui, Z., Veldhuis, R.: Toward face biometric de-identification using adversarial exam- ples. In: IEEE Conf. on Computers, Software, and Applications (COMPSAC). pp. 723–728 (June 2023), also presented...

    2023

  12. [19]

    In: 2024 IEEE International Conference on Image Processing Challenges and Workshops (ICIPCW)

    Ghafourian, M., Fierrez, J., Vera-Rodriguez, R., Tolosana, R., Morales, A.: SaFL: Sybil-aware federated learning with application to face recognition. In: 2024 IEEE International Conference on Image Processing Challenges and Workshops (ICIPCW). pp. 4050–4056 (2024)

    2024

  13. [20]

    arXiv:2302.10883 (2025)

    Ghafourian, M., Vera-Rodriguez, R., Fierrez, J., Sumer, B., Tolosana, R., Moralez, A., Kindt, E.: Blockchain and biometrics: Survey, GDPR elements, and future directions. arXiv:2302.10883 (2025)

    2025

  14. [21]

    IEEE Access 5, 8606–8619 (June 2017)

    Gomez-Barrero, M., Galbally, J., Morales, A., Fierrez, J.: Privacy-preserving com- parison of variable-length data with application to biometric template protection. IEEE Access 5, 8606–8619 (June 2017)

    2017

  15. [22]

    IEEE Access10 (Febru- ary 2022)

    Hassanpour,A.,Moradikia,M.,Yang,B.,Abdelhadi,A.,Busch,C.,Fierrez,J.:Dif- ferential privacy preservation in robust continual learning. IEEE Access10 (Febru- ary 2022)

    2022

  16. [23]

    arXiv:2310.06825 (2023)

    Jiang, A.Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D.S., Casas, D.d.l., Bressand, F., Lengyel, G., Lample, G., Saulnier, L., et al.: Mistral 7b. arXiv:2310.06825 (2023)

    2023 arXiv

  17. [24]

    Journal of Sys- tems and Software104, 82–89 (2015)

    Jurado, F., Rodriguez, P.: Sentiment analysis in monitoring software development processes: An exploratory case study on GitHub’s project issues. Journal of Sys- tems and Software104, 82–89 (2015)

    2015

  18. [25]

    Journal of Network and Computer Applications171, 102807 (2020)

    Kaaniche, N., Laurent, M., Belguith, S.: Privacy enhancing technologies for solving the privacy-personalization paradox: Taxonomy and survey. Journal of Network and Computer Applications171, 102807 (2020)

    2020

  19. [26]

    In: Proceedings of naacL-HLT

    Kenton, J.D.M.W.C., Toutanova, L.K.: Bert: Pre-training of deep bidirectional transformers for language understanding. In: Proceedings of naacL-HLT. vol. 1, p. 2 (2019)

    2019

  20. [27]

    Clinical Nutrition48, 80– 89 (2025)

    Lacruz-Pleguezuelos, B., Bazán, G.X., Romero-Tapiador, S., et al.: AI4Food, a feasibility study for the implementation of automated devices in the nutritional advice and follow up within a weight loss intervention. Clinical Nutrition48, 80– 89 (2025)

    2025

  21. [28]

    IScience24(3) (2021) PBa-LLM: Privacy- and Bias-aware NLP using NER 17

    Lepri, B., Oliver, N., Pentland, A.: Ethical Machines: The Human-centric Use of Artificial Intelligence. IScience24(3) (2021) PBa-LLM: Privacy- and Bias-aware NLP using NER 17

    2021

  22. [29]

    IEEE Transactions on Knowledge and Data Engineering34(1), 50–70 (2022)

    Li, J., Sun, A., Han, J., Li, C.: A survey on deep learning for named entity recogni- tion. IEEE Transactions on Knowledge and Data Engineering34(1), 50–70 (2022)

    2022

  23. [30]

    Computers & Industrial Engineering149, 106854 (2020)

    Li, L., Fan, Y., Tse, M., Lin, K.Y.: A review of applications in federated learning. Computers & Industrial Engineering149, 106854 (2020)

    2020

  24. [31]

    arXiv:1907.11692 (2019)

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

    2019 arXiv

  25. [32]

    arXiv preprint arXiv:2503.07384 (2025)

    Mancera, G., DeAlcala, D., Fierrez, J., Tolosana, R., Morales, A.: Is my text in your ai model? gradient-based membership inference test applied to llms. arXiv preprint arXiv:2503.07384 (2025)

    2025 arXiv

  26. [33]

    In: Proceedings of the Australasian Language Technology Association Workshop

    Mollá, D., Van Zaanen, M., Smith, D.: Named entity recognition for question an- swering. In: Proceedings of the Australasian Language Technology Association Workshop. pp. 51–58 (2006)

    2006

  27. [34]

    IEEE Transactions on Pattern Analysis and Machine Intelligence43(6), 2158–2164 (2020)

    Morales, A., Fierrez, J., Vera-Rodriguez, R., Tolosana, R.: SensitiveNets: Learning agnostic representations with application to face images. IEEE Transactions on Pattern Analysis and Machine Intelligence43(6), 2158–2164 (2020)

    2020

  28. [35]

    Mumtaz, R.: Extracting Named Entities and Relations From Text for Populating Knowledge Graph. Ph.D. thesis, Capital University (2023)

    2023

  29. [36]

    ACM Computing Surveys (CSUR)54(1), 1–39 (2021)

    Nasar, Z., Jaffry, S.W., Malik, M.K.: Named entity recognition and relation ex- traction: State-of-the-art. ACM Computing Surveys (CSUR)54(1), 1–39 (2021)

    2021

  30. [37]

    Journal of Internet Services and Applications 10(1), 1 (2019)

    Neumann,G.K.,Grace,P.,Burns,D.,Surridge,M.:Pseudonymizationriskanalysis in distributed systems. Journal of Internet Services and Applications 10(1), 1 (2019)

    2019

  31. [38]

    Proceedings of Advances in Neural Information Processing Systems35, 27730–27744 (2022)

    Ouyang, L., Wu, J., Jiang, X., Almeida, D., et al.: Training Language Models to Follow Instructions with Human Feedback. Proceedings of Advances in Neural Information Processing Systems35, 27730–27744 (2022)

    2022

  32. [39]

    SN Computer Science4(5), 434 (2023)

    Peña,A.,Serna,I.,Morales,A.,Fierrez,J.,Ortega,A.,Herrarte,A.,Alcantara,M., Ortega-Garcia, J.: Human-centric multimodal machine learning: Recent advances and testbed on AI-based recruitment. SN Computer Science4(5), 434 (2023)

    2023

  33. [41]

    Information Fusion 108, 102398 (2024)

    Peña, A., Morales, A., Fierrez, J., Ortega-Garcia, J., Puente, I., Cordova, J., Cor- dova, G.: Continuous document layout analysis: Human-in-the-loop AI-based data curation, database, and evaluation in the domain of public affairs. Information Fusion 108, 102398 (2024)

    2024

  34. [42]

    In: IAPR Intl

    Peña, A., Morales, A., Fierrez, J., Serna, I., Ortega-Garcia, J., Puente, I., Cordova, J., Cordova, G.: Leveraging large language models for topic classification in the do- main of public affairs. In: IAPR Intl. Conf. on Document Analysis and Recognition Workshops (ICDARw). LN...

    2023

  35. [43]

    In: IEEE/CVF Conf

    Peña, A., Serna, I., Morales, A., Fierrez, J.: Bias in multimodal AI: Testbed for fair automatic recruitment. In: IEEE/CVF Conf. on Computer Vision and Pat- tern Recognition Workshops (CVPRw) (June 2020), also presented at ICML 2020 Workshop on Human-in-the-Loop Learning

    2020

  36. [44]

    In: ACM Intl

    Peña, A., Serna, I., Morales, A., Fierrez, J.: FairCVtest Demo: Understanding bias in multimodal learning with a testbed in fair automatic recruitment. In: ACM Intl. Conf. on Multimodal Interaction (ICMI). pp. 760–761 (October 2020), also presented at European Conf. on Compute...

    2020

  37. [45]

    Pfitzmann, A., Hansen, M.: A terminology for talking about privacy by data minimization: Anonymity, unlinkability, undetectability, unobservability, pseudonymity, and identity management (2010) 18 Mancera et al

    2010

  38. [46]

    In: Proceedings of the Conference on Empirical Methods in Natural Language Processing

    Ritter, A., Clark, S., Etzioni, O., et al.: Named entity recognition in tweets: an experimental study. In: Proceedings of the Conference on Empirical Methods in Natural Language Processing. pp. 1524–1534 (2011)

    2011

  39. [47]

    IEEE Commu- nications Magazine 32(9), 40–48 (1994)

    Sandhu, R.S., Samarati, P.: Access control: principle and practice. IEEE Commu- nications Magazine 32(9), 40–48 (1994)

    1994

  40. [48]

    In: Proceedings of the Confer- ence on Natural Language Learning at HLT-NAACL 2003

    Sang, E.F., De Meulder, F.: Introduction to the CoNLL-2003 shared task: Language-independent named entity recognition. In: Proceedings of the Confer- ence on Natural Language Learning at HLT-NAACL 2003. pp. 142–147 (2003)

    2003

  41. [49]

    NYUL rev.86, 1814 (2011)

    Schwartz, P.M., Solove, D.J.: The PII problem: Privacy and a new concept of personally identifiable information. NYUL rev.86, 1814 (2011)

    2011

  42. [50]

    In: AAAI Workshop on Artificial Intelligence Safety (SafeAI)

    Serna, I., DeAlcala, D., Morales, A., Fierrez, J., Ortega-Garcia, J.: IFBiD: Inference-free bias detection. In: AAAI Workshop on Artificial Intelligence Safety (SafeAI). CEUR-WS, vol. 3087 (February 2022)

    2022

  43. [51]

    Artificial Intelligence 305, 103682 (April 2022)

    Serna, I., Morales, A., Fierrez, J., Obradovich, N.: Sensitive loss: Improving accu- racy and fairness of face representations with discrimination-aware deep learning. Artificial Intelligence 305, 103682 (April 2022)

    2022

  44. [52]

    Decision Support Sys- tems 107, 78–87 (2018)

    Serrano-Silva, Y.O., Villuendas-Rey, Y., Yánez-Márquez, C.: Automatic feature weighting for improving financial decision support systems. Decision Support Sys- tems 107, 78–87 (2018)

    2018

  45. [53]

    arXiv preprint arXiv:2402.13846 (2024)

    Staab, R., Vero, M., Balunović, M., Vechev, M.: Large language models are ad- vanced anonymizers. arXiv preprint arXiv:2402.13846 (2024)

    2024 arXiv

  46. [54]

    Wisconsin International Law Journal34(2), 284–322 (2017)

    Stalla-Bourdillon,S.,Knight,A.:Anonymousdatav.personaldata-falsedebate:an eu perspective on anonymization, pseudonymization and personal data. Wisconsin International Law Journal34(2), 284–322 (2017)

    2017

  47. [55]

    In: European Conf

    Tello, J., de la Cruz, M., Ribeiro, T., Fierrez, J., Morales, A., Tolosana, R., Alonso, C.L., Ortega, A.: Symbolic AI (LFIT) for XAI to handle biases. In: European Conf. on Artificial Intelligence Workshops (ECAIw). CEUR-WS, vol. 3523 (Octo- ber 2023)

    2023

  48. [56]

    IEEE Trans

    Terhorst, P., Kolf, J.N., Huber, M., Kirchbuchner, F., Damer, N., Morales, A., Fierrez, J., Kuijper, A.: A comprehensive study on face recognition biases beyond demographics. IEEE Trans. on Technology and Society3(1), 16–30 (March 2022)

    2022

  49. [57]

    (2024), https://www.whitehouse.gov/briefing- room/presidential-actions/2024/10/24/

    The White House: Memorandum on Advancing the United States’ Lead- ership in Artificial Intelligence. (2024), https://www.whitehouse.gov/briefing- room/presidential-actions/2024/10/24/

    2024

  50. [58]

    arXiv:2302.13971 (2023)

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al.: LLaMA: Open and efficient foundation language models. arXiv:2302.13971 (2023)

    2023 arXiv

  51. [59]

    arXiv:2307.09288 (2023)

    Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bash- lykov, N., Batra, S., Bhargava, P., Bhosale, S., et al.: LLaMA 2: Open foundation and fine-tuned chat models. arXiv:2307.09288 (2023)

    2023 arXiv

  52. [60]

    Proceedings of Advances in Neural Infor- mation Processing Systems (2017)

    Vaswani, A.: Attention is all you need. Proceedings of Advances in Neural Infor- mation Processing Systems (2017)

    2017

  53. [61]

    Springer (2025)

    Veldhuis, R., et al.: Privacy and Security Matters in Biometric Technologies. Springer (2025)

    2025

  54. [62]

    Springer (2017)

    Voigt, P., Von dem Bussche, A.: The EU General Data Protection Regulation (GDPR) - A Practical Guide. Springer (2017)

    2017

  55. [63]

    Journal of Chemical Information and Modeling59(9), 3692–3702 (2019)

    Weston, L., Tshitoyan, V., Dagdelen, J., Kononova, O., Trewartha, A., Persson, K.A., Ceder, G., Jain, A.: Named entity recognition and normalization applied to large-scale information extraction from the materials science literature. Journal of Chemical Information and Modelin...

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.