Exploration and Evaluation of Bias in Cyberbullying Detection with Machine Learning

Andrew Root; Liam Jakubowski; Mounika Vanamala

REVIEW 2 major objections 5 minor 18 references

Exploration and Evaluation of Bias in Cyberbullying Detection with Machine Learning

T0 review · 2 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Cross-dataset tests show cyberbullying models lose 0.222 Macro F1

desk verdict A useful, honest cross-dataset evaluation with a debatable label mapping that weakens the headline number; the paper is worth refereeing. read the letter →

arxiv 2412.00609 v1 pith:RF5LXPEF submitted 2024-11-30 cs.LG

Andrew Root , Liam Jakubowski , Mounika Vanamala This is my paper · ORCID

classification cs.LG

keywords cyberbullyingdetectioncross-datasetevaluationdatasetbiaslabelmappingdynamicqueryexpansionMacroF1scoremachinelearninggeneralizationTwitterdatasets

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 tries to establish that cyberbullying detection models trained on one Twitter dataset do not transfer to other Twitter datasets, even when training uses strong anti-overfitting measures. It reports an average drop of 0.222 in Macro F1 when the best models from cross-validation are tested on unseen corpora, and argues the gap comes from differences in how cyberbullying is defined, how tweets are collected, and how labels are assigned. The paper also singles out Dynamic Query Expansion (DQE), an automated method that labels new tweets using keywords from already-labeled ones, as a source of bias that can make a model approximate the expansion algorithm instead of human judgment. If correct, the findings imply that reported cyberbullying detection accuracy on a single dataset overstates real-world usefulness.

What carries the argument

The load-bearing machinery is the cross-dataset evaluation protocol: stratified k-fold cross-validation selects the top ten models from two gradient-boosting classifiers per dataset, then each is retrained on its own dataset and tested on the other two, with the Macro F1 drop relative to cross-validation as the outcome. Alongside this, the paper defines Dynamic Query Expansion (DQE) as a function DQE(D_y) = (x̂, y) that pulls a new tweet x̂ from outside the dataset and immediately assigns it the label y of the tweets that generated the query. This formulation shows how automated expansion can silently replace the human labeling function H with the DQE rule, and it motivates the prediction that models trained on such data will fail on datasets labeled by H.

What would settle it

Re-run the six cross-dataset experiments after moving Dataset 2's Offensive class from Not Cyberbullying to Cyberbullying (or excluding it entirely). If the average Macro F1 drop falls well below 0.222, the headline result is an artifact of the label mapping; if the drop remains, the conclusion of corpus-level bias is supported.

Watch

Extended reading notes

Core claim

The central claim is that the usefulness of a cyberbullying model is dataset-bound: a model that scores well in cross-validation on its training corpus drops substantially when evaluated on another corpus. In six experiments pairing three Twitter-based datasets, the average Macro F1 drop is 0.222 and the average weighted F1 drop is 0.253, with the largest single Macro F1 drop at 0.3912. The paper argues this drop is evidence of bias in data collection and labeling, and it formalizes DQE as a function that both fetches and labels tweets in one step, so that a model trained on DQE-expanded data may be approximating DQE rather than the human labeling process.

Load-bearing premise

The load-bearing premise is that the three datasets can be compared after mapping each original label to Cyberbullying or Not Cyberbullying, in particular treating Dataset 2's Offensive tweets as Not Cyberbullying; if that mapping is wrong, the reported 0.222 drop could be an artifact of the relabeling rather than of dataset bias.

Editorial extensions

If this is right

Cyberbullying models should not be deployed on new platforms or populations without first being tested on a locally labeled sample; a strong within-dataset score is not evidence of real-world readiness.
Dataset curation choices, especially lexicon-based collection and automated expansion, need to be reported alongside accuracy numbers, because they determine whether the model learned human labeling or the collection rule.
Cross-dataset evaluation, not just held-out splits of one dataset, should become a standard reporting requirement for cyberbullying detection research.
Models trained on DQE-expanded data may be approximating the expansion algorithm instead of human labelers, so their apparent performance is not a measure of cyberbullying detection ability.
An average drop of 0.222 in Macro F1 means a detector that looks reliable in research settings can be materially worse in practice.

Reading between the lines

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

If the label mapping is the real driver of the drop, then the same experimental setup could be reused to quantify definitional disagreement: by remapping labels under alternative definitions and measuring how much the average drop changes, one could turn a negative result into a measurement of labeling bias.
The critique of DQE extends beyond cyberbullying to any semi-supervised or synthetic-data augmentation pipeline in NLP: the danger is not only keyword bias but that the model learns the labeling shortcut, so a testable extension is to compare models trained on augmented versus human-labeled subsets of the same corpus while controlling for size.
The OOV correlation results suggest that vocabulary mismatch is not the main cause of the drop; a direct next experiment would be to measure whether the drop persists when all three datasets are filtered to a common vocabulary, isolating label-definition effects from lexical effects.
A practical application of the finding is that dataset health can be scored by cross-dataset transferability, giving dataset curators a cheap proxy for annotation quality without additional human labeling.

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.

Desk Editor's Note

read the letter

Quick take: this is a solid, honest empirical demonstration that cyberbullying models trained on one public Twitter corpus lose a lot of Macro F1 on another, with code and data shipped. The headline 0.222 average drop is believable as a qualitative claim, but the specific number is weaker than it looks because of how Dataset 2's 'Offensive' labels were folded into 'Not Cyberbullying.'

What's actually new: not the general point—cross-dataset generalization failures are well known in NLP, and the paper says so itself. The contribution is the three-dataset measurement set and the explicit formalization of Dynamic Query Expansion (DQE) as a labeling function that a model might approximate instead of human labelers. That formalism is simple but useful, and it makes the data collection critique concrete. The OOV correlation analysis is a nice negative result; it rules out one simple explanation for the drop.

Soft spots, in order of importance. The label mapping. Section 2.2.2 maps Hate Speech to Cyberbullying and Offensive plus Non-Offensive to Not Cyberbullying. Davidson et al. treat Offensive as a separate category from hate speech, and offensive language can be part of cyberbullying. Because Dataset 2 appears in four of the six experiments, this choice directly drives the average drop. The authors admit the conversion is not the only feasible way, but they never test an alternative mapping. This is fixable, not fatal: even if you remapped Offensive as Cyberbullying, the drops are large across all pairs, so the qualitative conclusion would likely survive. But the specific 0.222 number needs either a sensitivity analysis or a caveat that it depends on a contested mapping.

Second, there are no error bars or significance tests on the cross-dataset drops. Table 1 gives averages but no variance across the ten models per experiment, and the abstract's 'significant' is informal. Minor relative to the effect size.

Third, the DQE contamination estimate in Section 2.1.1 is unverifiable: the hypergeometric calculation uses quantities that the original paper doesn't document. This doesn't affect the cross-dataset results.

Who should read this: anyone working on cyberbullying detection or cross-dataset evaluation of text classifiers. It's a good case study for a seminar. It deserves a real referee: the empirical work is reproducible and the label mapping issue is easily fixed in revision.

Recommendation: send to peer review with a request for sensitivity analysis on the label mapping and either error bars or a more modest claim.

Referee Report

2 major / 5 minor

Summary. The manuscript performs cross-dataset evaluation of machine-learning models for cyberbullying detection on three public Twitter corpora. The authors relabel all datasets into a binary Cyberbullying/Not Cyberbullying scheme, train CatBoost and XGBoost models with hyperparameter search and cross-validation, then test each dataset's best models on the other two datasets. They report an average Macro F1 drop of 0.222 across six experiments and interpret this as evidence that current datasets are only useful for same-corpus prediction. The paper also formulates a critique of automated dataset-expansion methods such as Dynamic Query Expansion (DQE).

Significance. If the headline drop is robust, the study provides a useful, concrete cautionary demonstration of dataset shift in cyberbullying detection, with implications for model deployment and dataset curation. The experimental protocol is simple and transparent, code is publicly available, and the authors are unusually honest about limitations, explicitly stating in Section 5 that they do not provide causal evidence linking labeling differences to the observed drop. However, the central quantitative claim currently rests on a contested relabeling of one dataset's 'Offensive' class, and on averages reported without any measure of variability or significance testing; both issues must be addressed before the claim can be considered established.

major comments (2)

[Section 2.2.2, Table 1] The mapping of Dataset 2's 'Offensive' class to Not Cyberbullying is a load-bearing modeling choice. Dataset 2 participates in four of the six experiments (Exp. 1, 3, 4, and 6), so this mapping directly influences the average drop of 0.222 reported in Section 4. The authors themselves note that the conversion is 'not the only feasible way,' and the original Davidson et al. scheme treats Offensive as a distinct category from both Hate Speech and Non-Offensive, not as non-cyberbullying. A sensitivity analysis is needed: for example, re-run the pipeline with Offensive mapped to Cyberbullying, or with Dataset 2 excluded, and report how the average drop changes. Without this, the cross-dataset conclusion cannot be separated from the relabeling artifact.
[Section 3.2, Table 1] The abstract and Section 4 describe the drop as 'significant,' but no measure of variability or statistical test is provided. Each experiment averages over the top 10 models, so the manuscript should report the standard deviation across those models, a paired test comparing cross-validation and cross-dataset Macro F1 per model (e.g., Wilcoxon signed-rank), or a bootstrap confidence interval. This is particularly important because Exp. 6 has a negative drop in F1 Weighted (-0.0108), demonstrating substantial experiment-level variability that the current reporting hides.

minor comments (5)

[Section 2.1.1] The claim that Dataset 1 is 'likely overtly tainted' by DQE is supported only by an expected value (25,358.20) with no confidence interval or variance estimate. Since the DQE hypothesis is not directly tested in the experiments, this language should be softened or supplemented with a sensitivity analysis.
[Introduction, reference [8]] The text cites 'Van Hee et al.' for a definition of cyberbullying, but reference [8] is Van Cleemput et al., 'The development of a serious game on cyberbullying.' Please correct the citation or use the intended source.
[Section 2.2.3] There is a typo: 'different different learning rates' should read 'different learning rates.'
[Section 4] The terms 'Macro Mean F1,' 'F1 Macro Mean,' 'Macro F1 Score,' and 'F1 Weighted' are used inconsistently. Please standardize terminology to match Table 1.
[Section 5] The admission that the paper does 'not explicitly provide evidence that differing labeling schemes or definitions affect the overall performance drop' is important and should be reflected in the abstract and introduction, which currently frame the drop as evidence of bias from definitions and collection methods.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the cross-dataset evaluation tests models on held-out corpora, and the average Macro F1 drop is an empirical result rather than a construction of the method.

full rationale

The paper's central claim is that models trained on one cyberbullying dataset lose performance when evaluated on another dataset, with an average Macro F1 drop of 0.222. This claim is tested empirically: the models are fit on one dataset, selected by within-dataset stratified cross-validation, and then evaluated on the other two datasets that were not used in fitting or model selection. No parameter is fitted to the cross-dataset target values, and no equation defines the reported drop in terms of the hypothesis. The label conversion in Section 2.2.2, including the mapping of Dataset 2's Offensive class to Not Cyberbullying, is a preprocessing choice that could affect the magnitude of the measured drop, and the authors explicitly acknowledge it is 'not the only feasible way' to convert labels; however, this is a validity or robustness concern rather than circularity, because the conversion is not derived from the outcome being predicted. The DQE discussion is a formalization and critique of an existing data-expansion method, not a fitted result or a renamed prediction. The paper also contains no load-bearing self-citations: the cited prior works are external datasets and methods, and the authors' own prior work is not used to justify the conclusion. The cross-dataset drop is therefore a genuine empirical finding that could, in principle, have contradicted the hypothesis, so no circular step reduces the derivation to its inputs by construction.

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

The central claim is an empirical measurement, not a derivation, so it has few fitted constants. The main unstated inputs are the experimental design choices and the semantic label mappings; no new entities are introduced.

free parameters (4)

Early-stopping validation split fraction = 0.10
Chosen by the authors; affects final model weights and therefore the measured drop.
Cross-validation fold count = 6
Chosen by the authors; affects model selection and variance.
Top-model selection count = 10
Authors select the 10 highest-CV-F1 models per classifier; this selection can inflate the apparent cross-dataset drop.
Classifier hyperparameters (learning rate, tree depth, vectorizer) = Not reported; selected by search
The full search space is only in code; exact chosen values are absent from the paper, so the drop cannot be independently recomputed from text.

assumptions (4)

domain assumption The three Twitter datasets can be meaningfully compared after collapsing all labels to Cyberbullying vs Not Cyberbullying.
Adopted in Section 2.2.2; in particular Dataset 2's Offensive class is treated as Not Cyberbullying, a semantic choice that could drive the cross-dataset drop.
domain assumption DQE's auto-assigned labels are not equivalent to human labels H, so training on DQE data makes models approximate DQE rather than H.
Section 2.2.1 argues this; if false, DQE contamination would not explain Dataset 1's poor cross-dataset performance.
standard math Hypergeometric expectation formula applies to the DQE sampling process with the undisclosed per-class counts.
Section 2.1.1 uses E[S] = sum n|S_l|/N_l; the formula is standard but its inputs are not reported.
domain assumption Macro F1 averaged over six stratified folds is a valid control baseline for generalization.
Section 2.2.3 uses CV Macro F1 as the control; this assumes CV performance is a fair measure of in-distribution ability.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Exploration and Evaluation of Bias in Cyberbullying Detection with Machine Learning." pith.science (2026). https://pith.science/paper/RF5LXPEF

@misc{pith2026241200609,
  author       = {Pith},
  title        = {Pith review of: Exploration and Evaluation of Bias in Cyberbullying Detection with Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RF5LXPEF}},
  note         = {Machine review of arXiv:2412.00609}
}

read the original abstract

It is well known that the usefulness of a machine learning model is due to its ability to generalize to unseen data. This study uses three popular cyberbullying datasets to explore the effects of data, how it's collected, and how it's labeled, on the resulting machine learning models. The bias introduced from differing definitions of cyberbullying and from data collection is discussed in detail. An emphasis is made on the impact of dataset expansion methods, which utilize current data points to fetch and label new ones. Furthermore, explicit testing is performed to evaluate the ability of a model to generalize to unseen datasets through cross-dataset evaluation. As hypothesized, the models have a significant drop in the Macro F1 Score, with an average drop of 0.222. As such, this study effectively highlights the importance of dataset curation and cross-dataset testing for creating models with real-world applicability. The experiments and other code can be found at https://github.com/rootdrew27/cyberbullying-ml.

Figures

Figures reproduced from arXiv: 2412.00609 by the authors.

**Figure 2.** Cross Validation Results 3.2 Cross-Dataset Evaluation Results [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Cross-Dataset Evaluation Results. The red dashed line indicates the average across all experiments. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 15 canonical work pages

[1]

When the timeline meets the pipeline: A survey on automated cyberbullying detection,

F. Elsafoury, S. Katsigiannis, Z. Pervez, and N. Ramzan, “When the timeline meets the pipeline: A survey on automated cyberbullying detection,” IEEE Access, vol. 9, pp. 103 541–103 563, 2021

work page 2021
[2]

Automated hate speech detection and the problem of offensive language,

T. Davidson, D. Warmsley, M. W. Macy, and I. Weber, “Automated hate speech detection and the problem of offensive language,” CoRR, vol. abs/1703.04009, 2017. [Online]. Available: http://arxiv.org/abs/1703.04009

arXiv 2017
[3]

Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection,

J. Wang, K. Fu, and C.-T. Lu, “Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection,” in 2020 IEEE International Conference on Big Data (Big Data) , 2020, pp. 1699–1708

work page 2020
[4]

ID-XCB: Data-independent Debiasing for Fair and Accurate Transformer-based Cyberbullying Detection

P. Yi and A. Zubiaga, “Id-xcb: Data-independent debiasing for fair and accurate transformer-based cyberbullying detection,” 2024. [Online]. Available: https://arxiv.org/abs/2402.16458

work page Pith review arXiv 2024
[5]

Accurate cyberbullying detection and prevention on social media,

A. Perera and P. Fernando, “Accurate cyberbullying detection and prevention on social media,”Procedia Computer Science, vol. 181, pp. 605–611, 2021, cENTERIS 2020 - International Conference on ENTERprise Information Systems / ProjMAN 2020 - International Conference on Project MANagement / HCist 2020 - International Conference on Health and Social Care Inf...

work page 2021
[6]

Investigating the role of swear words in abusive language detection tasks,

E. W. Pamungkas, V . Basile, and V . Patti, “Investigating the role of swear words in abusive language detection tasks,” Language Resources and Evaluation , vol. 57, no. 1, pp. 155–188, Mar 2023. [Online]. Available: https://doi.org/10.1007/s10579-022-09582-8

work page doi:10.1007/s10579-022-09582-8 2023
[7]

Using machine learning to detect cyberbullying,

K. Reynolds, A. Edwards, and L. Edwards, “Using machine learning to detect cyberbullying,” Proceedings - 10th International Conference on Machine Learning and Applications, ICMLA 2011 , vol. 2, 12 2011

work page 2011
[8]

The development of a serious game on cyberbullying: A concept test,

K. Van Cleemput, H. Vandebosch, K. Poels, S. Bastiaensens, A. Desmet, and I. De Bourdeaudhuij, “The development of a serious game on cyberbullying: A concept test,” in Cyberbullying: From theory to intervention , T. Völlink, F. Dehue, and C. Mcguckin, Eds. Taylor & Francis Group, 2016, pp. 93–109

work page 2016

Show all 18 references

[9]

Mean birds: Detecting aggression and bullying on twitter,

D. Chatzakou, N. Kourtellis, J. Blackburn, E. D. Cristofaro, G. Stringhini, and A. Vakali, “Mean birds: Detecting aggression and bullying on twitter,” CoRR, vol. abs/1702.06877, 2017. [Online]. Available: http://arxiv.org/abs/1702.06877

2017 arXiv
[10]

A large labeled corpus for online harassment research,

J. Golbeck, Z. Ashktorab, R. O. Banjo, A. Berlinger, S. Bhagwan, C. Buntain, P. Cheakalos, A. A. Geller, Q. Gergory, R. K. Gnanasekaran, R. R. Gunasekaran, K. M. Hoffman, J. Hottle, V . Jienjitlert, S. Khare, R. Lau, M. J. Martindale, S. Naik, H. L. Nixon, P. Ramachandran, K. ...

2017
[11]

Identification of hate speech in social media,

N. Ruwandika and A. Weerasinghe, “Identification of hate speech in social media,” in2018 18th International Conference on Advances in ICT for Emerging Regions (ICTer) , 2018, pp. 273–278

2018
[12]

Hateful symbols or hateful people? predictive features for hate speech detection on Twitter,

Z. Waseem and D. Hovy, “Hateful symbols or hateful people? predictive features for hate speech detection on Twitter,” in Proceedings of the NAACL Student Research Workshop, J. Andreas, E. Choi, and A. Lazaridou, Eds. San Diego, California: Association for Computational Linguis...

2016
[13]

Analysing cyberbullying using natural language processing by understanding jargon in social media,

B. Bhatia, A. Verma, Anjum, and R. Katarya, “Analysing cyberbullying using natural language processing by understanding jargon in social media,” CoRR, vol. abs/2107.08902, 2021. [Online]. Available: https://arxiv.org/abs/2107.08902

2021 arXiv
[14]

Effective hate-speech detection in twitter data using recurrent neural networks,

G. Pitsilis, H. Ramampiaro, and H. Langseth, “Effective hate-speech detection in twitter data using recurrent neural networks,” Applied Intelligence, vol. 48, p. in press., 12 2018

2018
[15]

M. P. Deisenroth, A. A. Faisal, and C. S. Ong, Mathematics for Machine Learning . Cambridge University Press, 2020

2020
[16]

What the f-measure doesn’t measure: Features, flaws, fallacies and fixes,

D. M. W. Powers, “What the f-measure doesn’t measure: Features, flaws, fallacies and fixes,” CoRR, vol. abs/1503.06410, 2015. [Online]. Available: http://arxiv.org/abs/1503.06410

2015 arXiv
[17]

Generalizability of machine learning models: Quantitative evaluation of three methodological pitfalls,

F. Maleki, K. L. Ovens, R. Gupta, C. Reinhold, A. Spatz, and R. Forghani, “Generalizability of machine learning models: Quantitative evaluation of three methodological pitfalls,” CoRR, vol. abs/2202.01337, 2022. [Online]. Available: https://arxiv.org/abs/2202.01337 8

2022 arXiv
[2020]

Available: https://www.sciencedirect.com/science/article/pii/S1877050921002507

[Online]. Available: https://www.sciencedirect.com/science/article/pii/S1877050921002507

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

When the timeline meets the pipeline: A survey on automated cyberbullying detection,

F. Elsafoury, S. Katsigiannis, Z. Pervez, and N. Ramzan, “When the timeline meets the pipeline: A survey on automated cyberbullying detection,” IEEE Access, vol. 9, pp. 103 541–103 563, 2021

work page 2021

[2] [2]

Automated hate speech detection and the problem of offensive language,

T. Davidson, D. Warmsley, M. W. Macy, and I. Weber, “Automated hate speech detection and the problem of offensive language,” CoRR, vol. abs/1703.04009, 2017. [Online]. Available: http://arxiv.org/abs/1703.04009

arXiv 2017

[3] [3]

Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection,

J. Wang, K. Fu, and C.-T. Lu, “Sosnet: A graph convolutional network approach to fine-grained cyberbullying detection,” in 2020 IEEE International Conference on Big Data (Big Data) , 2020, pp. 1699–1708

work page 2020

[4] [4]

ID-XCB: Data-independent Debiasing for Fair and Accurate Transformer-based Cyberbullying Detection

P. Yi and A. Zubiaga, “Id-xcb: Data-independent debiasing for fair and accurate transformer-based cyberbullying detection,” 2024. [Online]. Available: https://arxiv.org/abs/2402.16458

work page Pith review arXiv 2024

[5] [5]

Accurate cyberbullying detection and prevention on social media,

A. Perera and P. Fernando, “Accurate cyberbullying detection and prevention on social media,”Procedia Computer Science, vol. 181, pp. 605–611, 2021, cENTERIS 2020 - International Conference on ENTERprise Information Systems / ProjMAN 2020 - International Conference on Project MANagement / HCist 2020 - International Conference on Health and Social Care Inf...

work page 2021

[6] [6]

Investigating the role of swear words in abusive language detection tasks,

E. W. Pamungkas, V . Basile, and V . Patti, “Investigating the role of swear words in abusive language detection tasks,” Language Resources and Evaluation , vol. 57, no. 1, pp. 155–188, Mar 2023. [Online]. Available: https://doi.org/10.1007/s10579-022-09582-8

work page doi:10.1007/s10579-022-09582-8 2023

[7] [7]

Using machine learning to detect cyberbullying,

K. Reynolds, A. Edwards, and L. Edwards, “Using machine learning to detect cyberbullying,” Proceedings - 10th International Conference on Machine Learning and Applications, ICMLA 2011 , vol. 2, 12 2011

work page 2011

[8] [8]

The development of a serious game on cyberbullying: A concept test,

K. Van Cleemput, H. Vandebosch, K. Poels, S. Bastiaensens, A. Desmet, and I. De Bourdeaudhuij, “The development of a serious game on cyberbullying: A concept test,” in Cyberbullying: From theory to intervention , T. Völlink, F. Dehue, and C. Mcguckin, Eds. Taylor & Francis Group, 2016, pp. 93–109

work page 2016