Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Enhancing Privacy in the Early Detection of Sexual Predators Through Federated Learning and Differential Privacy

T0 review · 4 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 Privacy-preserving federated learning can flag online grooming without centralizing private messages.

desk verdict Useful application paper on FL+DP for early grooming detection, with a solid DP-FedAvg result but an overstated abstract and a mislabeled DP-SGD privacy number. read the letter →

arxiv 2501.12537 v2 pith:OIMEP3LN submitted 2025-01-21 cs.CL cs.CY

Khaoula Chehbouni , Martine De Cock , Gilles Caporossi , Afaf Taik , Reihaneh Rabbany , Golnoosh Farnadi This is my paper · ORCID
classification cs.CLcs.CY
keywords earlydetectionofsexualpredatorsfederatedlearningdifferentialprivacyonlinegroomingDP-FedAvgprivacy-preservingtextclassificationF-latencyslidingwindowevaluation
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 asks whether an early-warning system for online grooming can be trained without collecting children's private conversations on a central server. It builds a federated-learning pipeline in which each user's conversation stays on their own device, and adds differential privacy in three different places: noise on text embeddings, noise on local gradients, and noise on aggregated updates. Using the PANC dataset and the sliding-window early-detection task, it reports that the best private variant, DP-FedAvg, reaches an F-latency of 62 percent, close to the 64 percent of non-private federated learning at a privacy budget of $\epsilon=1$. The paper's central claim is that privacy guarantees and detection utility can coexist with only a slight utility reduction.

What carries the argument

The mechanism is federated averaging (FedAvg) wrapped in three differential-privacy layers: metric DP, which adds calibrated noise to the $[\text{CLS}]$ embeddings before local training; DP-SGD, which clips and noisies gradients during local training; and DP-FedAvg, which clips client updates and adds noise to the aggregated update on the server. The evaluation machinery is the sliding-window early-detection protocol of Vogt et al. (2021) with the latency-weighted F1 score, which penalizes late warnings and rewards detection early enough to intervene.

What would settle it

Run the DP-FedAvg model from this paper on a held-out collection of real grooming conversations that are not drawn from Perverted Justice, using the same 50-message sliding window and skepticism level; if the F-latency falls well below the reported 62%, the claimed privacy-utility balance does not transfer to actual victims.

Watch

Extended reading notes

Core claim

The core discovery is that federated learning with user-level differential privacy preserves the practical utility of early grooming detection: DP-FedAvg scores 62% F-latency and DP-SGD scores 61%, against 64% for federated learning without privacy and 63% for centralized training, at $(\epsilon=1, \delta=10^{-5})$. The paper also finds that perturbing raw embeddings (metric DP) costs more utility, reaching only 52%, and that enforcing a 1% false-positive rate lowers detection speed for all models but leaves the private models usable. It concludes that a server can orchestrate training without ever seeing raw messages, and that user-level guarantees are the appropriate protection for an application where one user's entire conversation history is sensitive.

Load-bearing premise

The whole utility story rests on the assumption that Perverted Justice conversations, in which adults pose as children, are close enough to real grooming that the measured F-latency carries over to deployment; the paper's own limitations section notes these chats lack overt persuasion and sexual extortion.

Editorial extensions

If this is right

  • A deployed system could run detection locally on a user's device and raise warnings without any raw message ever leaving the device.
  • User-level differential privacy protects an entire conversation history rather than individual messages, which matches the sensitivity of the online-grooming application.
  • Flagged conversations could be added to a public warm-up pool to expand scarce labeled data without exposing the reporter's identity.
  • The same federated pipeline could transfer to other early-risk text tasks such as cyberbullying or depression detection, as the paper itself suggests.

Reading between the lines

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

  • The reported numbers are benchmarked on Perverted Justice transcripts, so the 62% figure should be read as an upper bound for real deployment until the model is checked on actual victim conversations.
  • Adding secure aggregation or a trusted execution environment to DP-FedAvg would close the one gap the paper leaves open: a server that can see non-private updates.
  • Since DP-FedAvg provides user-level privacy with almost no utility loss, it is the variant best matched to this application; DP-SGD's instance-level guarantee protects individual messages but not the whole conversation history.
  • The 1% false-positive threshold reveals a hard trade: lowering false accusations measurably slows detection, so deployment choices should be tuned to the local cost of false alarms.
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 / 6 minor

Summary. The paper studies privacy-preserving frameworks for early detection of sexual predators (eSPD) in online chats, using federated learning (FL) combined with three differential privacy (DP) mechanisms: metric DP applied to embeddings, DP-SGD applied during local training, and DP-FedAvg applied to aggregated updates. Experiments on the PANC dataset with a logistic regression classifier on BERT embeddings are evaluated under Vogt et al.'s sliding-window early-detection protocol. The main quantitative result is that FL with DP-FedAvg reaches an F-latency of 62% and FL with DP-SGD reaches 61%, compared with 64% for non-private FL, at reported privacy levels of (\epsilon=1, \delta=10^{-5}). The authors conclude that privacy and utility can coexist with only a slight utility reduction. The paper includes implementation details in appendices and makes code publicly available.

Significance. If the results are reliable, the paper makes a useful practical contribution: it demonstrates that a real-world early-warning system for grooming can be trained in a federated manner with formal DP guarantees while retaining most of the utility of centralized training. The evaluation follows an established task protocol (Vogt et al. 2021), uses a realistic dataset, and compares three distinct DP integration strategies, which is informative for practitioners. The released code and thorough hyperparameter reporting are strengths. However, the headline claim about DP-SGD's privacy guarantee is not currently supported because of the unresolved composition of per-round privacy budgets, and the utility differences between the best private and non-private models are small enough that the absence of error bars is a genuine concern.

major comments (4)
  1. [§6.3 and Appendix C] The DP-SGD privacy budget is computed with a client-side moment accountant for each round of federated training, but the total privacy loss across rounds is not composed. The experimental setup samples 10% of 10,000 clients per round for 100 rounds, so the expected number of selections per client is about 10, and most clients are resampled. The paper itself acknowledges in Appendix C that "the privacy budget computed is not the total privacy budget for the implementation, since the same client can be resampled." Therefore the reported (\epsilon=1, \delta=10^{-5}) for the DP-SGD row in Table 2 is a per-round budget, not a total budget, and the abstract's claim that the evaluation "proves that privacy and utility can coexist" is overstated for this configuration. The authors should either compute and report a composed total budget (e.g., using a server-side moment accountant with amplification via sampling) or clearly relabel the result as a per-round budget and adjust the privacy claims accordingly.
  2. [Table 2] Table 2 does not state the privacy parameters for each row. The text mentions (\epsilon=1, \delta=10^{-5}) for DP-FedAvg and DP-SGD, and \eta=20 for metric DP, but the table itself only lists model names. Without these parameters, the reader cannot assess the privacy-utility trade-off or verify that the configurations are comparable. The table should include the privacy budget (\epsilon, \delta, or \eta) for each model, and also the 57% F-latency result for DP-SGD at \epsilon=0.50 mentioned in §6.3.
  3. [§6.2 and Table 1] The classification threshold for a 1% false positive rate is selected using the test set and the same test set is then used to report the resulting F1, precision, and F-latency. This is a form of test-set overfitting: the reported numbers at 1% FPR are optimistic and would likely degrade on unseen data. A held-out validation set or cross-validation should be used for threshold selection before evaluating on the test set. This issue affects the specific FPR-controlled results in Table 1 and Figure 8, although the qualitative finding that tighter thresholds reduce speed may still hold.
  4. [Tables 1 and 2] All reported metrics appear to be from single runs, with no error bars or repeated-seed variance. The central claim of "only a slight reduction in utility" rests on differences of 1–3 percentage points in F-latency (64% for FL, 62% for DP-FedAvg, 61% for DP-SGD), which may fall within run-to-run noise given the random client sampling and warm-up splitting. At minimum, the authors should report the mean and standard deviation over several seeds (or over the three warm-up splits already used) for the main comparisons to establish that the private models' utility is not statistically worse than the non-private FL baseline.
minor comments (6)
  1. [Abstract] The word "proves" is too strong for a single empirical study with a specific dataset and classifier; consider "demonstrates" or "suggests" to match the actual scope of the evidence.
  2. [§6.3] The text introduces the DP-SGD result at (\epsilon=0.50, \delta=10^{-5}) with an F-latency of 57%, but this value is not shown in Table 2; the table should include it for completeness.
  3. [Appendix C] The phrase "can further amplified by randomized check-ins" is ungrammatical; it should read "can be further amplified." Also, the sentence about the moment accountant on the server is confusing and should be rewritten to distinguish the per-round client-side accounting from the needed cross-round composition.
  4. [Table 2] The column headers "Privacy Protection / Model / Input / Output" are unclear; the intended meaning (which component receives DP protection) should be explained directly in the table caption or in a footnote.
  5. [Appendix C, Figure 6] The text claims that at \eta=5 it is not possible to recover any original embeddings, while at \eta=55 the inversion attack succeeds 99% of the time; the figure axes and the definition of "accuracy" of the attack should be labeled explicitly so the reader can interpret these numbers.
  6. [§8] The sentence "In such a system, the user will be able to give feedback on the model's prediction" is followed by a discussion of notifying a third party; the relationship between user feedback and third-party notification should be clarified to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported F-latency scores come from a held-out test set and the privacy algorithms are external standards.

full rationale

The paper is an empirical evaluation rather than a derivation, so the circularity patterns do not apply. The F-latency metric is computed from predictions on a held-out test split (Section 5, Appendix A), and the federated/differentially-private training algorithms (FedAvg, DP-SGD, DP-FedAvg, metric DP) are standard external mechanisms. The only tuned quantities are hyperparameters: the metric-DP noise level eta is selected by an inversion attack on training embeddings (Appendix C), and the DP-SGD clipping/noise/batch/epoch settings are chosen by random search; in both cases utility is then reported on the test set, so the utility numbers are not forced by construction. The paper itself discloses that the DP-SGD epsilon=1 is a per-round budget, not a composed total budget ("the privacy budget computed is not the total privacy budget for the implementation, since the same client can be resampled"), which is a privacy-accounting limitation rather than circular reasoning. Separate methodological concerns, such as tuning the 1% FPR threshold on the test set before re-evaluating on the same test set (Section 6.2) and the PANC/PJ data possibly not transferring to real grooming, affect external validity but do not make any claimed result equivalent to its inputs by construction. No load-bearing step reduces to a self-citation or to a fitted parameter renamed as a prediction.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central experiments rest on the PANC dataset, a public warm-up set, simulated clients, and standard DP mechanisms. No new theoretical entities are introduced. The main ledger items are the choice of eta, test-set threshold calibration, and the incomplete DP-SGD composition accounting.

free parameters (5)
  • Metric DP noise parameter eta = 20
    Chosen via embedding inversion attack: at eta=20 the attack recovers only 34% of training embeddings. This is a privacy-utility knob, not a formal DP budget.
  • Classification threshold for 1% false positive rate = Not reported, set per model
    Section 6.2 selects the threshold on the test set and then evaluates on the same test set, making the 1% FPR results optimistic.
  • DP-SGD hyperparameters (clipping norm, learning rate, batch size, local epochs) = Selected by random grid search, final values not reported
    The (epsilon=1, F-latency 61%) result depends on these choices; without final values, reproduction is harder.
  • DP-FedAvg hyperparameters (update clipping, noise multiplier, rounds, clients per round) = Selected by random grid search, final values not reported
    The (epsilon=1, F-latency 62%) result depends on these choices.
  • Warm-up data size and class balance = 10% of training data; 10 positive and 10 negative segments per client
    Hand-chosen to compensate for extreme non-IID data; the paper notes real deployment would need a public labeled source.
assumptions (6)
  • domain assumption Grooming examples in PANC (from Perverted Justice) are representative of real online grooming
    Load-bearing for real-world utility; the paper itself cites Schneevogt et al. showing PJ chats differ from real grooming, including overt persuasion and sexual extortion.
  • domain assumption A balanced public warm-up dataset of grooming and non-grooming examples is available to distribute to all clients
    The non-IID mitigation depends on this; the paper acknowledges in Appendix C that in a real scenario this would need a public source, but such labeled data is scarce.
  • domain assumption The federated client simulation (one positive user's conversation plus 10 randomly merged negative users per client) approximates real cross-device FL
    This deviates from real users and may make the FL optimization easier or harder in ways not measured; flagged in Appendix C.
  • domain assumption The server can be trusted in the DP-FedAvg setting
    Required for user-level privacy; if the server is malicious, non-private updates leak before noise is added.
  • domain assumption Pretrained BERT [CLS] features are sufficient for eSPD without fine-tuning
    The paper chooses logistic regression on BERT features for scalability, but Vogt et al. report higher F-latency with fine-tuned BERT; the FL utility could be limited by this choice.
  • domain assumption Per-client DP-SGD privacy accounting over multiple rounds gives a meaningful user-level privacy guarantee at the reported epsilon
    The paper reports epsilon=1 for DP-SGD in Table 2 but notes in Section 6.3 that the total budget is not computed when clients are resampled; the guarantee is therefore weaker than advertised.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Privacy in the Early Detection of Sexual Predators Through Federated Learning and Differential Privacy." pith.science (2026). https://pith.science/paper/OIMEP3LN

@misc{pith2026250112537,
  author       = {Pith},
  title        = {Pith review of: Enhancing Privacy in the Early Detection of Sexual Predators Through Federated Learning and Differential Privacy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OIMEP3LN}},
  note         = {Machine review of arXiv:2501.12537}
}
read the original abstract

The increased screen time and isolation caused by the COVID-19 pandemic have led to a significant surge in cases of online grooming, which is the use of strategies by predators to lure children into sexual exploitation. Previous efforts to detect grooming in industry and academia have involved accessing and monitoring private conversations through centrally-trained models or sending private conversations to a global server. In this work, we implement a privacy-preserving pipeline for the early detection of sexual predators. We leverage federated learning and differential privacy in order to create safer online spaces for children while respecting their privacy. We investigate various privacy-preserving implementations and discuss their benefits and shortcomings. Our extensive evaluation using real-world data proves that privacy and utility can coexist with only a slight reduction in utility.

Figures

Figures reproduced from arXiv: 2501.12537 by the authors.

Figure 1
Figure 1. Illustration of the three DP implementations de [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. illustrates the training phase of our framework. A global server selects clients to participate and distributes a model to them; the clients will then further train the model in a privacy-preserving manner on their mobile devices using their own personal data as well as a portion of warm-up data, as we can see in Alice’s cellular device [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 4
Figure 4. Illustration of the different privacy implementa [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figures from the paper (3 more)
Figure 3
Figure 3. Figure 3: eSPD: Inference Phase B.3 The Privacy Implementations B.4 The Deployed System In [PITH_FULL_IMAGE:figures/full_fig_p011_3.png]
Figure 6
Figure 6. Figure 6: Impact of the privacy budget η on the F-latency score of the FL model with metric DP and on the accuracy of the embeddings inversion attack. token embedding inversion (Qu et al. 2021b). We add noise η to each embeddings in our training set to obtain perturbed embedding…
Figure 7
Figure 7. Figure 7: Distribution of the warning latencies for the full [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Fairness in Federated Learning: Fairness for Whom?

    cs.LG 2025-05 conditional novelty 6.0 of 10

    A critical review of 121 federated learning fairness papers identifies five recurring pitfalls and proposes a harm-centered, lifecycle-based framework.

Reference graph

Works this paper leans on

70 extracted references · 62 canonical work pages · cited by 1 Pith paper

  1. [1]

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

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    B.; Mironov, I.; Talwar, K.; and Zhang, L

    Abadi, M.; Chu, A.; Goodfellow, I.; McMahan, H. B.; Mironov, I.; Talwar, K.; and Zhang, L. 2016. Deep Learning with Differential Privacy. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security . ACM

    work page 2016

  4. [4]

    Adams, S.; Melanson, D.; and De Cock , M. 2021. Private Text Classification with Convolutional Neural Networks. In Proceedings of the Third Workshop on Privacy in Natural Language Processing, 53--58

    work page 2021

  5. [5]

    A.; and Bours, P

    Agarwal, N.; \" U nl\" u , T.; Wani, M. A.; and Bours, P. 2021. Predatory Conversation Detection Using Transfer Learning Approach. In Machine Learning, Optimization, and Data Science: 7th International Conference, LOD 2021, Grasmere, UK, October 4–8, 2021, Revised Selected Papers, Part I, 488–499. Berlin, Heidelberg: Springer-Verlag. ISBN 978-3-030-95466-6

    work page 2021

  6. [6]

    Balle, B.; Kairouz, P.; McMahan, B.; Thakkar, O.; and Guha Thakurta, A. 2020. Privacy amplification via random check-ins. Advances in Neural Information Processing Systems, 33: 4623--4634

    work page 2020

  7. [7]

    S.; Naidu, R.; and Muftuoglu, Z

    Basu, P.; Roy, T. S.; Naidu, R.; and Muftuoglu, Z. 2021. Privacy enabled Financial Text Classification using Differential Privacy and Federated Learning. In Proceedings of the Third Workshop on Economics and Natural Language Processing, 50--55. Punta Cana, Dominican Republic: Association for Computational Linguistics

    work page 2021

  8. [8]

    J.; Topal, T.; Mathur, A.; Qiu, X.; Parcollet, T.; and Lane, N

    Beutel, D. J.; Topal, T.; Mathur, A.; Qiu, X.; Parcollet, T.; and Lane, N. D. 2020. Flower: A Friendly Federated Learning Research Framework. CoRR, abs/2007.14390

    arXiv 2020

Show all 70 references
  1. [9]

    Brown, H.; Lee, K.; Mireshghallah, F.; Shokri, R.; and Tramer, F. 2022. What Does it Mean for a Language Model to Preserve Privacy? In 2022 ACM Conference on Fairness, Accountability, and Transparency, 2280--2292

    2022

  2. [10]

    Canada, P. S. 2024. About Online Child Sexual Exploitation

    2024

  3. [11]

    Carlini, N.; Ippolito, D.; Jagielski, M.; Lee, K.; Tramer, F.; and Zhang, C. 2022. Quantifying Memorization Across Neural Language Models

    2022

  4. [12]

    Carlini, N.; Liu, C.; Erlingsson, \'U .; Kos, J.; and Song, D. 2019. The secret sharer: Evaluating and testing unintended memorization in neural networks. In 28th USENIX Security Symposium , 267--284

    2019

  5. [13]

    Chee, F. Y. 2023. EU lawmakers back rules forcing Big Tech to tackle child pornography

    2023

  6. [14]

    Collier, K.; and Burke, M. 2022. Facebook turned over chat messages between mother and daughter now charged over abortion

    2022

  7. [15]

    Devlin, J.; Chang, M.-W.; Lee, K.; and Toutanova, K. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv:1810.04805

    2019 arXiv

  8. [16]

    Dwork, C. 2008. Differential Privacy: A Survey of Results. In Agrawal, M.; Du, D.; Duan, Z.; and Li, A., eds., Theory and Applications of Models of Computation, 1--19. Berlin, Heidelberg: Springer Berlin Heidelberg. ISBN 978-3-540-79228-4

    2008

  9. [17]

    L.; et al

    Errecalde, M. L.; et al. 2017. Temporal Variation of Terms as Concept Space for Early Risk Prediction. In Clef (working notes)

    2017

  10. [18]

    J.; Montes y Gomez, M.; Villasenor, L.; and Errecalde, M

    Escalante, H. J.; Montes y Gomez, M.; Villasenor, L.; and Errecalde, M. L. 2016. Early text classification: a Na \" ve solution. In Proceedings of the 7th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, 91--99. San Diego, California: ...

    2016

  11. [19]

    European Commission . 2022. Fighting child sexual abuse: Commission proposes new rules to protect children

    2022

  12. [20]

    European Parliament . 2024. Child sexual abuse online: current rules extended until April 2026

    2024

  13. [21]

    Feyisetan, O.; Balle, B.; Drake, T.; and Diethe, T. 2020. Privacy- and Utility-Preserving Textual Analysis via Calibrated Multivariate Perturbations. In Proceedings of the 13th International Conference on Web Search and Data Mining, WSDM '20, 178–186. New York, NY, USA: Associ...

    2020

  14. [22]

    Fredrikson, M.; Jha, S.; and Ristenpart, T. 2015. Model inversion attacks that exploit confidence information and basic countermeasures. In Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security, 1322--1333

    2015

  15. [23]

    Goggin, B. 2023. Meta defaults Facebook and Messenger to end-to-end encryption despite child safety concerns

    2023

  16. [24]

    Government of Canada . 2023 a . Overrepresentation of Black People in the Canadian Criminal Justice System

    2023

  17. [25]

    Government of Canada . 2023 b . Overrepresentation of Indigenous People in the Canadian Criminal Justice System: Causes and Responses

    2023

  18. [26]

    Hill, K. 2012. Yes, Facebook Scans Users' Private Conversations Looking For Sexual Predators and Child Porn

    2012

  19. [27]

    Hill, K. 2022. A Dad Took Photos of His Naked Toddler for the Doctor. Google Flagged Him as a Criminal

    2022

  20. [28]

    R.; Zec, E

    Hilmkil, A.; Callh, S.; Barbieri, M.; S \"u tfeld, L. R.; Zec, E. L.; and Mogren, O. 2021. Scaling Federated Learning for Fine-Tuning of Large Language Models. In M \'e tais, E.; Meziane, F.; Horacek, H.; and Kapetanios, E., eds., Natural Language Processing and Information Sy...

    2021

  21. [29]

    Hu, Y.; Li, D.; Tan, Z.; Li, X.; and Wang, J. 2021. Adaptive Clipping Bound of Deep Learning with Differential Privacy. In 2021 IEEE 20th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom), 428--435

    2021

  22. [30]

    Inches, G.; and Crestani, F. 2012. Overview of the International Sexual Predator Identification Competition at PAN -2012. In CLEF (Online working notes/labs/workshop), volume 30

    2012

  23. [31]

    B.; Avent, B.; Bellet, A.; Bennis, M.; Bhagoji, A

    Kairouz, P.; McMahan, H. B.; Avent, B.; Bellet, A.; Bennis, M.; Bhagoji, A. N.; Bonawitz, K. A.; Charles, Z.; Cormode, G.; Cummings, R.; D'Oliveira, R. G. L.; Rouayheb, S. E.; Evans, D.; Gardner, J.; Garrett, Z.; Gasc \' o n, A.; Ghazi, B.; Gibbons, P. B.; Gruteser, M.; Harcha...

    2019 arXiv

  24. [32]

    Kairouz, P.; Oh, S.; and Viswanath, P. 2015. The composition theorem for differential privacy. In International conference on machine learning, 1376--1385. PMLR

    2015

  25. [33]

    Klymenko, O.; Meisenbacher, S.; and Matthes, F. 2022. Differential Privacy in Natural Language Processing The Story So Far. In Proceedings of the Fourth Workshop on Privacy in Natural Language Processing

    2022

  26. [34]

    Lee, G.; et al. 2022. Privacy-Preserving Text Classification on BERT Embeddings with Homomorphic Encryption. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 3169--3175. Seattle, ...

    2022

  27. [35]

    H.; de Gusm\ a o, P

    Li, K. H.; de Gusm\ a o, P. P. B.; Beutel, D. J.; and Lane, N. D. 2021. Secure Aggregation for Federated Learning in Flower. In Proceedings of the 2nd ACM International Workshop on Distributed Machine Learning, DistributedML '21, 8–14. New York, NY, USA: Association for Comput...

    2021

  28. [36]

    P.; et al

    Liang, P. P.; et al. 2021. Towards understanding and mitigating social biases in language models. In International Conference on Machine Learning, 6565--6576. PMLR

    2021

  29. [37]

    Liu, D.; et al. 2017. A Novel Way of Identifying Cyber Predators. CoRR, abs/1712.03903

    2017 arXiv

  30. [38]

    P.; Gonz \'a lez, F

    L \'o pez-Monroy, A. P.; Gonz \'a lez, F. A.; Montes, M.; Escalante, H. J.; and Solorio, T. 2018. Early Text Classification Using Multi-Resolution Concept Representations. In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computational...

    2018

  31. [39]

    Lorenzo-Dus, N.; et al. 2016. Understanding grooming discourse in computer-mediated environments. Discourse, Context & Media, 12: 40--50

    2016

  32. [40]

    E.; et al

    Losada, D. E.; et al. 2020. Overview of ERisk 2020: Early Risk Prediction on the Internet. In Experimental IR Meets Multilinguality, Multimodality, and Interaction: 11th International Conference of the CLEF Association, CLEF 2020, Thessaloniki, Greece, September 22–25, 2020, P...

    2020

  33. [41]

    Mattern, J.; Weggenmann, B.; and Kerschbaum, F. 2022. The Limits of Word Level Differential Privacy. In Carpuat, M.; de Marneffe, M.-C.; and Meza Ruiz, I. V., eds., Findings of the Association for Computational Linguistics: NAACL 2022, 867--881. Seattle, United States: Associa...

    2022

  34. [42]

    Mcghee, I.; Bayzick, J.; Edwards, A.; Edwards, L.; Mcbride, A.; and Jakubowski, E. 2011. Learning to Identify Internet Sexual Predation. International Journal of Electronic Commerce, 15

    2011

  35. [43]

    McMahan, B.; Moore, E.; Ramage, D.; Hampson, S.; and y Arcas, B. A. 2017 a . Communication-efficient learning of deep networks from decentralized data. In Artificial Intelligence and Statistics, 1273--1282. PMLR

    2017

  36. [44]

    B.; Ramage, D.; Talwar, K.; and Zhang, L

    McMahan, H. B.; Ramage, D.; Talwar, K.; and Zhang, L. 2017 b . Learning Differentially Private Language Models Without Losing Accuracy. CoRR, abs/1710.06963

    2017 arXiv

  37. [45]

    Morgan, J.; Paiement, A.; Lorenzo-Dus, N.; Kinzel, A.; and Cristofaro, M. D. 2021. Integrating linguistic knowledge into DNN s: Application to online grooming detection

    2021

  38. [46]

    Olson, L.; Daggs, J.; Ellevold, B.; and Rogers, T. 2007. Entrapping the innocent: Toward a theory of child sexual predators’ luring communication. Communication Theory, 17(3): 231--251

    2007

  39. [47]

    Popescu, M.; and Grozea, C. 2012. Kernel Methods and String Kernels for Authorship Analysis. In CLEF (Online Working Notes/Labs/Workshop). Citeseer

    2012

  40. [48]

    Qu, C.; Kong, W.; Yang, L.; Zhang, M.; Bendersky, M.; and Najork, M. 2021 a . Natural Language Understanding with Privacy-Preserving BERT. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, CIKM '21, 1488–1497. New York, NY, USA: Ass...

    2021

  41. [49]

    Qu, C.; Kong, W.; Yang, L.; Zhang, M.; Bendersky, M.; and Najork, M. 2021 b . Privacy-Adaptive BERT for Natural Language Understanding. CoRR, abs/2104.07504

    2021 arXiv

  42. [50]

    Radauskas, Gintaras . 2023. EU’s plans to combat online child abuse pose risk to encryption – interview

    2023

  43. [51]

    Reich, D.; Todoki, A.; Dowsley, R.; De Cock , M.; and Nascimento, A. 2019. Privacy-Preserving Classification of Personal Text Messages with Secure Multi-Party Computation. In Advances in Neural Information Processing Systems (NeurIPS), volume 32, 3752--3764

    2019

  44. [52]

    Resende, A.; Railsback, D.; Dowsley, R.; Nascimento, A. C. A.; and Aranha, D. F. 2022. Fast Privacy-Preserving Text Classification Based on Secure Multiparty Computation. IEEE Transactions on Information Forensics and Security, 17: 428--442

    2022

  45. [53]

    Rezaee Borj, P.; Raja, K.; and Bours, P. 2023. Detecting Online Grooming By Simple Contrastive Chat Embeddings. In Proceedings of the 9th ACM International Workshop on Security and Privacy Analytics, IWSPA '23, 57–65. New York, NY, USA: Association for Computing Machinery. ISB...

    2023

  46. [54]

    Sadeque, F.; et al. 2018. Measuring the Latency of Depression Detection in Social Media. In Proceedings of the Eleventh ACM International Conference on Web Search and Data Mining, WSDM '18, 495–503. New York, NY, USA: Association for Computing Machinery. ISBN 9781450355810

    2018

  47. [55]

    A.; Khan, U.; Khan, S.; Jamjoom, M

    Samee, N. A.; Khan, U.; Khan, S.; Jamjoom, M. M.; Sharif, M.; and Kim, D. H. 2023. Safeguarding Online Spaces: A Powerful Fusion of Federated Learning, Word Embeddings, and Emotional Features for Cyberbullying Detection. IEEE Access

    2023

  48. [56]

    Schneevogt, D.; Chiang, E.; and Grant, T. 2018. Do Perverted Justice chat logs contain examples of overt persuasion and sexual extortion? A research note responding to Chiang and Grant (2017, 2018). Language and Law/Linguagem e Direito, 5(1): 97--102

    2018

  49. [57]

    P.; Muniyal, B.; Priyanshu, A.; and Das, V

    Shetty, N. P.; Muniyal, B.; Priyanshu, A.; and Das, V. R. 2023. FedBully: A Cross-Device Federated Approach for Privacy Enabled Cyber Bullying Detection using Sentence Encoders. Journal of Cyber Security and Mobility, 12(04): 465–496

    2023

  50. [58]

    Shokri, R.; Stronati, M.; Song, C.; and Shmatikov, V. 2017. Membership Inference Attacks Against Machine Learning Models. In 2017 IEEE Symposium on Security and Privacy (SP), 3--18

    2017

  51. [59]

    Snowden, E. 2021. The all-seeing "I": Apple just declared war on your privacy

    2021

  52. [60]

    Song, C.; Ristenpart, T.; and Shmatikov, V. 2017. Machine learning models that remember too much. In Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, 587--601

    2017

  53. [61]

    D.; et al

    Thakkar, O. D.; et al. 2021. Understanding unintended memorization in language models under federated learning. In Proceedings of the Third Workshop on Privacy in Natural Language Processing, 1--10

    2021

  54. [62]

    K.; and Ristenpart, T

    Tram \`e r, F.; Zhang, F.; Juels, A.; Reiter, M. K.; and Ristenpart, T. 2016. Stealing machine learning models via prediction APIs . In 25th USENIX Security Symposium, 601--618

    2016

  55. [63]

    J.; Montes-y G \'o mez, M.; and Pineda, L

    Villatoro-Tello, E.; Ju \'a rez-Gonz \'a lez, A.; Escalante, H. J.; Montes-y G \'o mez, M.; and Pineda, L. V. 2012. A Two-step Approach for Effective Detection of Misbehaving Users in Chats. In CLEF (Online Working Notes/Labs/Workshop)

    2012

  56. [64]

    Vincent, J. 2022. New EU rules would require chat apps to scan private messages for child abuse. The Verge

    2022

  57. [65]

    Vogt, M.; et al. 2021. Early Detection of Sexual Predators in Chats. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 4985--4999. Onlin...

    2021

  58. [66]

    O.; Sarp, S.; Kuzlu, M.; and Cali, U

    Wibawa, F.; Catak, F. O.; Sarp, S.; Kuzlu, M.; and Cali, U. 2022. Homomorphic Encryption and Federated Learning based Privacy-Preserving CNN Training: COVID-19 Detection Use-Case

    2022

  59. [67]

    Zambrano, P.; et al. 2019. Technical Mapping of the Grooming Anatomy Using Machine Learning Paradigms: An Information Security Approach. IEEE Access, 7: 142129--142146

    2019

  60. [68]

    Zhao, Y.; et al. 2018. Federated Learning with Non-IID Data. CoRR, abs/1806.00582

    2018 arXiv

  61. [69]

    Zhu, H.; Xu, J.; Liu, S.; and Jin, Y. 2021. Federated Learning on Non-IID Data: A Survey. CoRR, abs/2106.06843

    2021 arXiv

  62. [70]

    Zhu, X.; Wang, J.; Hong, Z.; and Xiao, J. 2020. Empirical studies of institutional federated learning for natural language processing. In Findings of the Association for Computational Linguistics: EMNLP 2020, 625--634

    2020

Pith tools

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