Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Mario at EXIST 2025: A Simple Gateway to Effective Multilingual Sexism Detection

T0 review · 4 major / 5 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 The paper claims that a parameter-efficient LoRA fine-tune of Llama 3.1 8B, with adapters routed by predicted parent labels and one bilingual training run, took first place on all three EXIST 2025 sexism subtasks while training only 1.67%…

desk verdict First-place EXIST 2025 result with a simple LoRA recipe; the routing gimmick is unproven but the bilingual training gains and efficiency numbers are solid. read the letter →

arxiv 2507.10996 v1 pith:6CCR44DP submitted 2025-07-15 cs.CL

Lin Tian , Johanne R. Trippas , Marian-Andrei Rizoiu This is my paper · ORCID
classification cs.CL
keywords SexismDetectionLow-RankAdaptationHierarchicalClassificationSocialMediaAnalysisMultilingualLlama3.1Parameter-EfficientFine-TuningEXIST2025
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 top-tier multilingual sexism detection can come from a deliberately simple pipeline: one parameter-efficient fine-tune instead of ensembles, task-specific full models, or elaborate preprocessing. The system places first on all three EXIST 2025 Task 1 text subtasks across English and Spanish, and the authors attribute this to hierarchical LoRA adapters routed by parent-label predictions plus joint bilingual training. They show that the shared model beats separate language-specific models by 1.7–2.4 F1 points, and that adapting every linear layer rather than only attention helps. If these results stand, state-of-the-art performance on this kind of hierarchically structured moderation task no longer demands heavy resources.

What carries the argument

The carrying mechanism is hierarchical Low-Rank Adaptation (LoRA), defined by $W = W_0 + BA$: the pretrained weights $W_0$ stay frozen and small trainable matrices $A, B$ of rank 16 are added. Unlike usual LoRA recipes, the paper adapts every linear layer, including feed-forward projections and the language-modeling head. Each of the three subtasks gets its own adapter, and routing is conditional: at level $\ell$ the representation is $f_\theta(x) + \Delta^{(\ell)}_{\hat{y}^{(\ell-1)}}(x)$, so the adapter is chosen by the label predicted at the level above. A hierarchy loss $\mathcal{L}_{\text{hierarchy}}$ penalizes high-confidence child predictions when the parent is NOT_SEXIST, and the total objective combines the three task losses with this constraint.

What would settle it

Train the identical three adapters as a flat multi-task LoRA, with no parent-conditioned routing and no hierarchy loss, on the same English-Spanish data, and score both variants on the EXIST 2025 test set; if the flat model matches or beats the routed model on subtasks 1.2 and 1.3, the routing claim collapses. A second check is to feed gold parent labels at inference instead of the model's own predictions and see whether test scores improve.

Watch

Extended reading notes

Core claim

The central claim, stated on the paper's own terms, is that a simple hierarchical LoRA fine-tune of Llama 3.1 8B is enough to finish first on all three EXIST 2025 Task 1 text subtasks. Concretely, the system reports hard-label ICM-Hard scores of 0.6774 for binary sexism identification, 0.4991 for source intention detection, and 0.6519 for sexism categorization, all with one model covering English and Spanish. The authors credit three choices: applying LoRA to all linear transformations, routing adapters conditionally on parent-label predictions, and unified bilingual training that outperforms separate per-language models by 1.7–2.4 F1 points. Training uses only 1.67% of the model's parameters, cutting training time by 75% and per-adapter storage by 98% compared with full fine-tuning.

Load-bearing premise

The method's advantage rests on the untested assumption that routing a tweet to the next subtask's adapter according to its predicted parent label helps the child subtasks, even though training uses gold parent labels and no comparison against a flat model without routing is given.

Editorial extensions

If this is right

  • State-of-the-art results on this shared task were achievable with a single LoRA fine-tune, not an ensemble or a language-specific pipeline.
  • A single bilingual model outperforms separate English and Spanish models by 1.7–2.4 F1 points, removing the need for per-language systems.
  • Training uses only 1.67% of the model's parameters, 12GB of GPU memory, and a quarter of the full-fine-tuning time, with per-subtask adapters of 512MB.
  • Conditional adapter routing plus a hierarchy-consistency loss lets one base model handle three nested tasks without task-specific full models.

Reading between the lines

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

  • The paper never ablates the routing mechanism against a flat multi-task LoRA trained on all three subtasks, so the hierarchy's contribution to the scores is untested; a flat model might perform just as well.
  • Because training routes adapters with gold parent labels while inference routes with the model's own predicted labels, errors at the binary level can cascade into subtasks 1.2 and 1.3; an oracle-ablation that feeds true parents at inference would quantify this loss.
  • The hierarchy-loss weight $\lambda$ is never reported, leaving the actual trained objective underspecified and making exact reproduction harder than it appears.
  • The large Spanish-over-English gap on subtasks 1.2 and 1.3 may reflect annotation or label-distribution differences rather than any intrinsic property of Spanish sexist discourse; per-class error analysis would settle it.
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 / 5 minor

Summary. The paper describes the authors' submission to EXIST 2025 Task 1, which asks systems to detect sexism in English and Spanish tweets across three hierarchical subtasks: binary sexism identification (1.1), source intention detection (1.2), and sexism type categorization (1.3). The method uses QLoRA fine-tuning of Llama 3.1 8B with rank-16 adapters applied to all linear layers, trained jointly on English and Spanish data. The distinctive component is conditional adapter routing: separate LoRA adapters are trained for each hierarchical level, and at inference the adapter for a child subtask is selected based on the parent subtask's predicted label. A soft hierarchy loss penalizes confident child-level predictions for inputs predicted NOT_SEXIST at the binary level. The paper reports first place on all three subtasks on the EXIST 2025 hard-label test set (ICM-Hard scores 0.6774, 0.4991, and 0.6519), validation-set improvements from joint bilingual training over language-specific models (1.7--2.4 F1 points), and efficiency gains of 98.33% fewer trainable parameters relative to full fine-tuning.

Significance. If the central claims hold, the paper is a useful practical demonstration that a simple, parameter-efficient, unified multilingual model can be highly competitive in a shared task that is usually dominated by larger ensembles or language-specific systems. The test-set scores are externally anchored to the EXIST 2025 benchmark, the cross-lingual gains are reported consistently across all three subtasks, and the efficiency figures are concrete and plausible. However, the paper's distinctive scientific contribution is the hierarchical label-aware routing mechanism, and this mechanism is never isolated: no ablation compares it with a flat multi-task LoRA model or with non-hierarchical adapter selection. The exact training objective is also underspecified because the hierarchy loss weight lambda is never reported. These omissions mean the first-place scores support the overall pipeline but do not currently establish that the routing mechanism itself provides the claimed benefit. The work is therefore a reasonable working-notes contribution whose central methodological claim needs additional experiments to be fully supported.

major comments (4)
  1. [Section 3.3, Table 3] The central contribution, conditional adapter routing, is never ablated. The test-set results in Table 3 are obtained with the full pipeline, so they cannot show whether routing adapters based on the parent label improves Subtasks 1.2 and 1.3 relative to a flat multi-task LoRA model, a single shared adapter, or three independently trained adapters. Since Section 4.3 claims that the results 'validate our design choice of conditional LoRA routing,' an ablation that removes the routing mechanism (or replaces it with an unweighted mixture) is required to support that claim. I would expect this comparison on the validation set and, if possible, on the test set via the official evaluation.
  2. [Section 3.3, Eq. for L_hierarchy] The training objective is underspecified because the hierarchy constraint weight lambda is never reported in Table 1 or elsewhere. The paper defines L_hierarchy as an additive term in L_total, but without the value of lambda the exact loss cannot be reproduced. Additionally, the paper trains routing on gold parent labels but applies it at inference using predicted parent labels, and it does not specify what the child adapter produces for items predicted NOT_SEXIST, for which no Subtask 1.2/1.3 gold labels exist. The authors should state the value of lambda, clarify how child-level training instances are selected given the hierarchical label structure, and explain the inference behavior when the parent prediction is NOT_SEXIST.
  3. [Section 4.2, Table 3] The claim that the system 'achieved first place across all three subtasks, showing the effectiveness of hierarchical LoRA adaptation' is too strong. The ranking is an external result, but the causal attribution to the routing mechanism is not established by Table 3, which reports only the authors' own scores. A comparison with at least the runner-up systems or a simple non-hierarchical baseline is needed before the first-place result can be read as evidence for the routing design specifically. The current text makes this attribution in Section 4.2, but no data in the paper isolates the routing component.
  4. [Table 5] The rank ablation in Table 5 is reported only for Subtask 1.1 on the validation set and has no error bars or repeated runs. The F1 differences between rank 16 (0.868), rank 32 (0.869), and rank 64 (0.871) are within the range where random seed variation could matter, so the conclusion that 'rank 16 achieves optimal efficiency-performance trade-offs' is reasonable as a practical choice but is not strongly supported as a finding about model behavior. A brief statement about run-to-run variance, or a small number of seeds, would make the efficiency argument more robust.
minor comments (5)
  1. [Section 3.2] The text says 'our experiments showed that module targeting yields better performance' but no experiment comparing attention-only LoRA with all-linear LoRA is reported; either cite a result or present the comparison.
  2. [Table 2] The validation F1 improvements of 1.6--2.4 points are reported without significance tests, confidence intervals, or seed variation; given the small margins, the claim of 'consistent improvements' should be tempered or supported.
  3. [Section 4.3, Table 4] The training-time comparison (24 hours vs. 6 hours per task) would be more interpretable if the hardware, number of epochs, and early-stopping criterion were stated; the current presentation makes the 4x speedup claim hard to verify.
  4. [References] Several reference entries contain duplicated or malformed URLs (e.g., references [5], [35], and [39]); these should be cleaned up before publication.
  5. [Section 3.3] The symbol y-hat is used both for predicted labels in the hierarchy loss and for the routing condition, but the paper later says training routing uses gold parent labels; the notation should distinguish gold-parent-conditioned routing from predicted-parent-conditioned routing to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's results are anchored to the external EXIST 2025 test set, its ablations are empirical comparisons, and its self-citations are contextual rather than load-bearing.

full rationale

The paper reports scores on an external shared-task test set (EXIST 2025) using the official ICM-Hard metric, so the central 'first place' claims are not derived from the paper's own fitted values or definitions. The multilingual gain is supported by a stated validation-set ablation comparing joint bilingual training against separate language-specific models (Table 2), and the LoRA rank selection is supported by a separate rank ablation (Table 5); neither experiment defines the outcome in terms of the input. The hierarchical routing mechanism is admittedly not isolated by an ablation, so the paper's attribution of test-set success to that mechanism is under-supported empirically, but this is an evidential gap rather than a circular dependency: the reported numbers would remain the same even if the routing contributed nothing. Self-citations (e.g., references [8], [9], [36], [37]) appear as background, motivation, or future-work connections, and none is invoked as an unverified premise that forces the reported results; nor does the paper import a uniqueness theorem or define a quantity in terms of the prediction it claims to derive. There is no equation in which an output is identical by construction to a fitted input, and no fitted parameter is renamed as a prediction. Accordingly, no circularity step can be exhibited, and the appropriate finding is no significant circularity.

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

The central empirical claims rest on standard LoRA assumptions plus the EXIST benchmark. Key untested hyperparameters are rank and lambda. No new physical or conceptual entities are introduced.

free parameters (4)
  • LoRA rank r = 16
    Chosen via validation ablation; higher ranks (32, 64) give only <0.3% F1 gains with 2-4x parameters (Table 5).
  • LoRA alpha = 16
    Set equal to rank; no ablation reported.
  • Learning rate = 2e-4
    Chosen with constant schedule and 10% warmup; no sensitivity analysis.
  • Hierarchy constraint weight lambda = not reported
    Appears in the total loss (Section 3.3, Eq. 1) but its value is never stated and its contribution is not ablated.
assumptions (4)
  • standard math Low-rank update W = W0 + BA can approximate full fine-tuning for this task.
    Assumed from LoRA literature; not proven here.
  • domain assumption EXIST 2025 gold labels are correct ground truth for sexism detection.
    The model is trained and evaluated on these labels; no error analysis of the gold labels is provided.
  • ad hoc to paper Llama 3.1 8B has native bilingual capabilities that enable cross-lingual transfer.
    The paper relies on this to justify joint training; it is not independently established beyond the observed validation gains.
  • ad hoc to paper Conditioning at inference on the predicted parent label approximates the gold-label conditioning used in training.
    Section 3.3 uses gold labels during training but predicted labels at inference; error propagation is not analyzed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mario at EXIST 2025: A Simple Gateway to Effective Multilingual Sexism Detection." pith.science (2026). https://pith.science/paper/6CCR44DP

@misc{pith2026250710996,
  author       = {Pith},
  title        = {Pith review of: Mario at EXIST 2025: A Simple Gateway to Effective Multilingual Sexism Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6CCR44DP}},
  note         = {Machine review of arXiv:2507.10996}
}
read the original abstract

This paper presents our approach to EXIST 2025 Task 1, addressing text-based sexism detection in English and Spanish tweets through hierarchical Low-Rank Adaptation (LoRA) of Llama 3.1 8B. Our method introduces conditional adapter routing that explicitly models label dependencies across three hierarchically structured subtasks: binary sexism identification, source intention detection, and multilabel sexism categorization. Unlike conventional LoRA applications that target only attention layers, we apply adaptation to all linear transformations, enhancing the model's capacity to capture task-specific patterns. In contrast to complex data processing and ensemble approaches, we show that straightforward parameter-efficient fine-tuning achieves strong performance. We train separate LoRA adapters (rank=16, QLoRA 4-bit) for each subtask using unified multilingual training that leverages Llama 3.1's native bilingual capabilities. The method requires minimal preprocessing and uses standard supervised learning. Our multilingual training strategy eliminates the need for separate language-specific models, achieving 1.7-2.4\% F1 improvements through cross-lingual transfer. With only 1.67\% trainable parameters compared to full fine-tuning, our approach reduces training time by 75\% and model storage by 98\%, while achieving competitive performance across all subtasks (ICM-Hard: 0.6774 for binary classification, 0.4991 for intention detection, 0.6519 for multilabel categorization).

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. AI Wizards at EXIST 2026: Hierarchical Soft-Label Learning for Multimodal Sexism Identification in Memes

    cs.CL 2026-07 accept novelty 4.0 of 10

    Hierarchical gated-MLP soft-label multi-task learning on frozen Gemini embeddings ranked 1st on EXIST 2026 Task 2.3 and 4th on Tasks 2.1–2.2 Soft-Soft.

Reference graph

Works this paper leans on

42 extracted references · 33 canonical work pages · cited by 1 Pith paper

  1. [1]

    Talat, D

    Z. Talat, D. Hovy, Hateful symbols or hateful people? predictive features for hate speech detection on twitter, in: Proceedings of the NAACL student research workshop, 2016, pp. 88–93

    work page 2016

  2. [2]

    Davidson, D

    T. Davidson, D. Warmsley, M. Macy, I. Weber, Automated hate speech detection and the problem of offensive language, in: Proceedings of the international AAAI conference on web and social media, volume 11, 2017, pp. 512–515

    2017

  3. [3]

    Bailo, A

    F. Bailo, A. Johns, M.-A. Rizoiu, Riding information crises: the performance of far-right twitter users in australia during the 2019–2020 bushfires and the covid-19 pandemic, Information, Com- munication & Society (2023) 1–19. URL: https://www.tandfonline.com/doi/full/10.1080/1369118X. 2023.2205479. doi:10.1080/1369118X.2023.2205479

    work page arXiv 2023

  4. [4]

    Q. Kong, E. Booth, F. Bailo, A. Johns, M.-A. Rizoiu, Slipping to the extreme: A mixed method to explain how extreme opinions infiltrate online discussions, Proceedings of the International AAAI Conference on Web and Social Media 16 (2022) 524–535. URL: https://ojs.aaai.org/index. php/ICWSM/article/view/19312. doi:10.1609/icwsm.v16i1.19312

    work page doi:10.1609/icwsm.v16i1.19312 2022

  5. [5]

    P. J. Schneider, M.-A. Rizoiu, The effectiveness of moderating harmful online content, Proceedings of the National Academy of Sciences 120 (2023) 1–3. URL: https://www.pnas.org/doi/10.1073/pnas. 2307360120https://pnas.org/doi/10.1073/pnas.2307360120. doi:10.1073/pnas.2307360120

    doi:10.1073/pnas 2023

  6. [6]

    L. Yuan, P. J. Schneider, M.-A. Rizoiu, Behavioral homophily in social media via inverse re- inforcement learning: A reddit case study, Proceedings of the International Web Conference (WWW) (2025). URL: http://arxiv.org/abs/2502.02943http://dx.doi.org/10.1145/3696410.3714618. doi:10.1145/3696410.3714618

    work page Pith review arXiv 2025

  7. [7]

    Plaza, J

    L. Plaza, J. Carrillo-de Albornoz, I. Arcos, P. Rosso, D. Spina, E. Amigó, J. Gonzalo, R. Morante, Exist 2025: Learning with disagreement for sexism identification and characterization in tweets, memes, and tiktok videos, in: European Conference on Information Retrieval, Springer, 2025, pp. 442–449

    work page 2025

  8. [8]

    L. Yuan, T. Wang, G. Ferraro, H. Suominen, M.-A. Rizoiu, Transfer learning for hate speech detection in social media, Journal of Computational Social Science 6 (2023) 1081–1101. URL: https://link.springer.com/10.1007/s42001-023-00224-9. doi:10.1007/s42001-023-00224-9

    work page doi:10.1007/s42001-023-00224-9 2023

Show all 42 references
  1. [9]

    Yuan, M.-A

    L. Yuan, M.-A. Rizoiu, Generalizing hate speech detection using multi-task learning: A case study of political public figures, Computer Speech & Language 89 (2025) 101690. URL: https: //linkinghub.elsevier.com/retrieve/pii/S0885230824000731. doi:10.1016/j.csl.2024.101690

    2025

  2. [10]

    E. J. Hu, yelong shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, W. Chen, LoRA: Low-rank adaptation of large language models, in: International Conference on Learning Representations, 2022

    2022

  3. [11]

    Grattafiori, A

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

    2024 arXiv

  4. [12]

    Rodríguez-Sánchez, J

    F. Rodríguez-Sánchez, J. Carrillo-de Albornoz, L. Plaza, Automatic classification of sexism in social networks: An empirical study on twitter data, IEEE Access 8 (2020) 219563–219576

    2020

  5. [13]

    Rodríguez-Sánchez, J

    F. Rodríguez-Sánchez, J. Carrillo-de Albornoz, L. Plaza, A. Mendieta-Aragón, G. Marco-Remón, M. Makeienko, M. Plaza, J. Gonzalo, D. Spina, P. Rosso, Overview of exist 2022: sexism identification in social networks, Procesamiento del Lenguaje Natural 69 (2022) 229–240

    2022

  6. [14]

    Plaza, J

    L. Plaza, J. Carrillo-de Albornoz, R. Morante, E. Amigó, J. Gonzalo, D. Spina, P. Rosso, Overview of exist 2023: sexism identification in social networks, in: European Conference on Information Retrieval, Springer, 2023, pp. 593–599

    2023

  7. [15]

    H. Kirk, W. Yin, B. Vidgen, P. Röttger, Semeval-2023 task 10: Explainable detection of online sexism, in: Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023), 2023, pp. 2193–2210

    2023

  8. [16]

    Nobata, J

    C. Nobata, J. Tetreault, A. Thomas, Y. Mehdad, Y. Chang, Abusive language detection in online user content, in: Proceedings of the 25th international conference on world wide web, 2016, pp. 145–153

    2016

  9. [17]

    Badjatiya, S

    P. Badjatiya, S. Gupta, M. Gupta, V. Varma, Deep learning for hate speech detection in tweets, in: Proceedings of the 26th international conference on World Wide Web companion, 2017, pp. 759–760

    2017

  10. [18]

    Zhang, L

    Z. Zhang, L. Luo, Hate speech detection: A solved problem? the challenging case of long tail on twitter, Semantic Web 10 (2019) 925–945

    2019

  11. [19]

    Devlin, M.-W

    J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, Bert: Pre-training of deep bidirectional transformers for language understanding, in: Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, v...

    2019

  12. [20]

    E. W. Pamungkas, V. Basile, V. Patti, Misogyny detection in twitter: a multilingual and cross-domain study, Information processing & management 57 (2020) 102360

    2020

  13. [21]

    call me sexist, but

    M. Samory, I. Sen, J. Kohne, F. Flöck, C. Wagner, “call me sexist, but... ”: Revisiting sexism detection using psychological scales and adversarial samples, in: Proceedings of the international AAAI conference on web and social media, volume 15, 2021, pp. 573–584

    2021

  14. [22]

    Chiril, E

    P. Chiril, E. W. Pamungkas, F. Benamara, V. Moriceau, V. Patti, Emotionally informed hate speech detection: a multi-target perspective, Cognitive Computation (2022) 1–31

    2022

  15. [23]

    Mikolov, I

    T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, J. Dean, Distributed representations of words and phrases and their compositionality, Advances in neural information processing systems 26 (2013)

    2013

  16. [24]

    Pennington, R

    J. Pennington, R. Socher, C. D. Manning, Glove: Global vectors for word representation, in: Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), 2014, pp. 1532–1543

    2014

  17. [25]

    D. Q. Nguyen, T. Vu, A.-T. Nguyen, Bertweet: A pre-trained language model for english tweets, in: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, 2020, pp. 9–14

    2020

  18. [26]

    Barbieri, J

    F. Barbieri, J. Camacho-Collados, L. E. Anke, L. Neves, Tweeteval: Unified benchmark and comparative evaluation for tweet classification, in: Findings of the Association for Computational Linguistics: EMNLP 2020, 2020, pp. 1644–1650

    2020

  19. [27]

    Conneau, K

    A. Conneau, K. Khandelwal, N. Goyal, V. Chaudhary, G. Wenzek, F. Guzmán, É. Grave, M. Ott, L. Zettlemoyer, V. Stoyanov, Unsupervised cross-lingual representation learning at scale, in: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 202...

    2020

  20. [28]

    D. Nozza, Exposing the limits of zero-shot cross-lingual hate speech detection, in: Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th Inter- national Joint Conference on Natural Language Processing (Volume 2: Short Papers), 20...

    2021

  21. [29]

    who’s gonna take care of your children when you are at acl?

    P. Chiril, V. Moriceau, F. Benamara, A. Mari, G. Origgi, M. Coulomb-Gully, He said “who’s gonna take care of your children when you are at acl?”: Reported sexist acts are not sexist, in: 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020), ACL: Asso...

    2020

  22. [30]

    Guest, B

    E. Guest, B. Vidgen, A. Mittos, N. Sastry, G. Tyson, H. Margetts, An expert annotated dataset for the detection of online misogyny, in: Proceedings of the 16th conference of the European chapter of the association for computational linguistics: main volume, 2021, pp. 1336–1350

    2021

  23. [31]

    T. K. Smith, H. R. Nie, J. R. Trippas, D. Spina, RMIT-IR at EXIST Lab at CLEF 2024, in: Working Notes of CLEF 2024 – Conference and Labs of the Evaluation Forum, 2024

    2024

  24. [32]

    K.-L. Chiu, A. Collins, R. Alexander, Detecting hate speech with gpt-3, arXiv preprint arXiv:2103.12407 (2021)

    2021 arXiv

  25. [33]

    Brown, B

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, et al., Language models are few-shot learners, Advances in neural information processing systems 33 (2020) 1877–1901

    2020

  26. [34]

    W. Yin, A. Zubiaga, Towards generalisable hate speech detection: a review on obstacles and solutions, PeerJ Computer Science 7 (2021) e598

    2021

  27. [35]

    Q. Kong, P. Calderon, R. Ram, O. Boichak, M.-A. Rizoiu, Interval-censored transformer hawkes: Detecting information operations using the reaction of social systems, in: Proceedings of the ACM Web Conference 2023, ACM, 2023, pp. 1813–1821. URL: http://arxiv.org/abs/2211.14114ht...

    2023 arXiv

  28. [36]

    L. Tian, E. Booth, F. Bailo, J. Droogan, M.-A. Rizoiu, Before it’s too late: A state space model for the early prediction of misinformation and disinformation engagement, in: Proceedings of the International Web Conference (WWW), 2025. doi:10.1145/3696410.3714527

    2025

  29. [37]

    Tian, M.-A

    L. Tian, M.-A. Rizoiu, Estimating online influence needs causal modeling! counterfactual analysis of social media engagement (2025)

    2025

  30. [38]

    Calderon, R

    P. Calderon, R. Ram, M.-A. Rizoiu, Opinion market model: Stemming far-right opinion spread using positive interventions, Proceedings of the International AAAI Conference on Web and Social Media 18 (2024) 177–190. URL: https://ojs.aaai.org/index.php/ICWSM/article/view/31306. do...

    2024 doi

  31. [39]

    R. Ram, E. Thomas, D. Kernot, M.-A. Rizoiu, Detecting extreme ideologies in shifting landscapes: an automatic & context-agnostic approach, in: International AAAI Conference on Web and Social Media (ICWSM), AAAI, 2025. URL: http://arxiv.org/abs/2208.04097

    2025 arXiv

  32. [40]

    Dao, Flashattention-2: Faster attention with better parallelism and work partitioning, in: The Twelfth International Conference on Learning Representations, 2024

    T. Dao, Flashattention-2: Faster attention with better parallelism and work partitioning, in: The Twelfth International Conference on Learning Representations, 2024

    2024

  33. [41]

    Amigo, A

    E. Amigo, A. Delgado, Evaluating extreme hierarchical multi-label classification, in: Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022, pp. 5809–5819

    2022

  34. [42]

    A. F. Magnosão de Paula, J. S. Culpepper, A. Moffat, S. P. Cherumanal, F. Scholer, J. Trippas, The Effects of Demographic Instructions on LLM Personas, in: Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ‘25, 2025

    2025

Pith tools

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