The rise of user-generated content on social media has led to increasing instances of toxic speech—such as hate, abuse, and cyberbullying—which threaten online safety and integrity. Manual moderation struggles to cope with the vast content volume and the mental burden it places on human moderators, highlighting the urgent need for automated detection systems. However, many existing approaches require large labeled datasets, which are costly and difficult to obtain. To address this challenge, we propose U-GIFT (Uncertainty-Guided Few-Shot Detection of Toxic Speech), a self-training framework that excels in low-resource settings. U-GIFT integrates Bayesian neural networks with an uncertainty-guided sample selection strategy to identify high-confidence pseudo-labeled samples from unlabeled data. This strategy improves detection accuracy even with minimal supervision. Experiments across multiple benchmarks show that U-GIFT outperforms strong baselines, achieving a 14.92% gain in the 5-shot setting. The method can be applied to different pre-trained language models, is robust to sample imbalance, and generalizes well across domains. By reducing dependence on annotated data, U-GIFT offers an effective, scalable solution for toxic speech detection and supports safer online interactions through improved automated moderation.
VogelsEA. The state of online harassment. Pew Res Center2021.
2.
VedelerJSOlsenTEriksenJ. Hate speech harms: A social justice discussion of disabled norwegians’ experiences. Disabil Soc2019; 34: 368–383.
3.
BleichE. The rise of hate speech and hate crime laws in liberal democracies. J Ethn Migr Stud2011; 37: 917–934.
4.
GillespieT. Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media. New Haven, CT: Yale University Press, 2018.
5.
ChenYZhouYZhuS. Detecting offensive language in social media to protect adolescent online safety. In: 2012 International Conference on Privacy, Security, Risk and Trust / 2012 International Conference on Social Computing, 2012, pp.71–80, IEEE.
6.
HuttoCGilbertE. VADER: A parsimonious rule-based model for sentiment analysis of social media text. Proc Int AAAI Conf Web Soc Med2014; 8: 216–225.
7.
ElSheriefMKulkarniVNguyenD, et al.Hate speech detection in social media. In: Proceedings of the International AAAI conference on web and social media Vol. 12, 2018.
8.
KarayiğitHAcıÇIAkdağıA. Detecting abusive instagram comments in turkish using convolutional neural network and machine learning methods. Expert Syst Appl2021; 174: 114802.
9.
ZhangZRobinsonDTepperJ. Detecting hate speech on Twitter using a convolution-GRU based deep neural network. In: The Semantic Web, 2018, pp.745–760. Springer.
10.
DevlinJChangM-WLeeK, et al.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. 2019, pp.4171–4186.
MozafariMFarahbakhshRCrespiN. A BERT-based transfer learning approach for hate speech detection in online social media. In: Complex networks and their applications VIII, 2020, pp.928–940. Springer.
13.
MnassriKRajapakshaPFarahbakhshR, et al.BERT-based ensemble approaches for hate speech detection. In: GLOBECOM 2022 – IEEE global communications conference, 2022, pp.4649–4654. IEEE.
14.
WiedemannGYimamSMBiemannC. UHH-LT at SemEval-2020 Task 12: Fine-tuning of pre-trained transformer networks for offensive language detection. In: Proceedings of the fourteenth workshop on semantic evaluation. 2020, pp.1638–1644.
15.
PolettoFBasileVSanguinettiM, et al.Resources and benchmark corpora for hate speech detection: A systematic review. Lang Resour Eval2021; 55: 477–523.
16.
ChenL-MXiuB-XDingZ-Y. Multiple weak supervision for short text classification. Appl Intell2022; 52: 9101–9116.
17.
DuttaPChakrabortySRoychowdhuryS, et al.CRUSH: Contextually regularized and user anchored self-supervised hate speech detection. Findings of NAACL2022: 1874–1886.
18.
AlsafariSSadaouiS. Semi-supervised self-training of hate and offensive speech from social media. Appl Artif Intell2021; 35: 1621–1645.
19.
LudwigFDolosKAlves-PintoA, et al.Unraveling the dynamics of semi-supervised hate speech detection: The impact of unlabeled data characteristics and pseudo-labeling strategies. In: Findings of the association for computational linguistics: EACL 2024. 2024, pp.1974–1986.
20.
MozafariMFarahbakhshRCrespiN. Cross-lingual few-shot hate speech and offensive language detection using meta learning. IEEE Access2022; 10: 14880–14896.
21.
SahaPShethDKediaK, et al.Rationale-guided few-shot classification to detect abusive language. In: ECAI 2023, 2023, pp.2041–2048. IOS Press.
22.
FinnCAbbeelPLevineS. Model-agnostic meta-learning for fast adaptation of deep networks. In: International conference on machine learning. 2017, pp.1126–1135.
23.
LeeD-H. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In: Workshop on challenges in representation learning, ICML. 2013, pp.896.
24.
WaseemZHovyD. 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.
25.
MalmasiSZampieriM. Challenges in discriminating profanity from hate speech. J Exp Theor Artif Intell2018; 30: 187–202.
26.
ChatzakouDKourtellisNBlackburnJ, et al.Measuring GamerGate: A tale of hate, sexism, and bullying. In: Proceedings of the 26th international conference on world wide web companion. 2017, pp.1285–1290.
27.
DavidsonTWarmsleyDMacyM, et al.Automated hate speech detection and the problem of offensive language. In: Proceedings of the International AAAI conference on web and social media. 2017, pp.512–515.
28.
MacAvaneySYaoH-RYangE, et al.Hate speech detection: Challenges and solutions. PLoS ONE2019; 14: 1–16.
29.
MahajanEMahajanHKumarS. EnsMulHateCyb: Multilingual hate speech and cyberbully detection in online social media. Expert Syst Appl2024; 236: 121228.
30.
VaswaniAShazeerNParmarN, et al.Attention is all you need. Adv Neural Inf Process Syst2017; 30.
31.
KennedyBJinXDavaniAM, et al.Contextualizing hate speech classifiers with post-hoc explanation. In: Proceedings of the 58th Annual meeting of the association for computational linguistics. 2020, pp.5435–5442.
32.
MouGYePLeeK. SWE2: Subword enriched and significant word emphasized framework for hate speech detection. In: Proceedings of the 29th ACM international conference on information & knowledge management. 2020, pp.1145–1154.
33.
TekiroluSSChungY-LGueriniM. Generating counter narratives against online hate speech: Data and strategies. In: Proceedings of the 58th annual meeting of the association for computational linguistics. 2020, pp.1177–1190.
34.
RadfordAWuJChildR, et al.Language models are unsupervised multitask learners. OpenAI Blog2019; 1: 9.
35.
ZhouXYongYFanX, et al.Hate speech detection based on sentiment knowledge sharing. In: Proceedings of the 59th annual meeting of the association for computational linguistics. 2021, pp.7158–7166.
36.
Rodríguez-SánchezFCarrillo-de-AlbornozJPlazaL. Detecting sexism in social media: An empirical analysis of linguistic patterns and strategies. Appl Intell2024; 54: 10995–11019.
37.
FirminoAABaptistaCSPaivaAC. Improving hate speech detection using cross-lingual learning. Expert Syst Appl2024; 235: 121115.
38.
VinyalsOBlundellCLillicrapT, et al.Matching networks for one-shot learning. Adv Neural Inf Process Syst2016; 29.
39.
SnellJSwerskyKZemelR. Prototypical networks for few-shot learning. Adv Neural Inf Process Syst2017; 30.
40.
AlKhamissiBLadhakFIyerS, et al.ToKen: Task decomposition and knowledge infusion for few-shot hate speech detection. In: Proceedings of the 2022 conference on empirical methods in natural language processing. 2022, pp.2109–2120.
41.
HeKPuNLaoM, et al.Dual selective knowledge transfer for few-shot classification. Appl Intell2023; 53: 27779–27789.
42.
AminiM-RFeofanovVPaulettoL, et al.Self-training: A survey. Neurocomputing2025; 616: 128904.
43.
LiDZhuXSongL. Mutual match for semi-supervised online evolutive learning. Appl Intell2023; 53: 3336–3350.
44.
ZhaoBWangYSuK, et al.Semi-supervised pedestrian re-identification via a teacher–student model with similarity-preserving generative adversarial networks. Appl Intell2023; 53: 1605–1618.
45.
WanJLiXZhaoJ, et al.Joint uncertainty model and metric for robust feature selection: A bi-level distribution consideration and feature evaluation approach. Fuzzy Sets Syst2026; 523: 109615.
46.
GalYGhahramaniZ. Dropout as a Bayesian approximation: Representing model uncertainty in deep learning. In: International conference on machine learning. PMLR, 2016.
47.
GawlikowskiJRovile NjieutcheuTCAliM, et al.A survey of uncertainty in deep neural networks. Artif Intell Rev2023; 56: 1513–1589.
48.
MukherjeeSAwadallahA. Uncertainty-aware self-training for few-shot text classification. Adv Neural Inf Process Syst2020; 33: 21199–21212.
49.
HavasiMJenattonRFortS, et al.Training independent subnetworks for robust prediction. arXiv preprint arXiv:2010.06610, 2020.
50.
DurasovNBagautdinovTBaqueP, et al.Masksembles for uncertainty estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021.
51.
SensoyMKaplanLKandemirM. Evidential deep learning to quantify classification uncertainty. Adv Neural Inf Process Syst2018; 31.
52.
HeidrichSBeemelmannsTNekrasovA, et al.OCCUQ: Exploring efficient uncertainty quantification for 3D occupancy prediction. arXiv preprint arXiv:2503.10605, 2025.
53.
GanaieMAHuMMalikAK, et al.Ensemble deep learning: A review. Eng Appl Artif Intell2022; 115: 105151.
54.
HeJZhangXLeiS, et al.CLUR: Uncertainty estimation for few-shot text classification with contrastive learning. In: Proceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining. 2023.
MathewBSahaPYimamSM, et al.Hatexplain: A benchmark dataset for explainable hate speech detection. Proc AAAI Conf Artif Intell2021; 35: 14867–14875.
57.
JiangAYangXLiuY, et al.SWSR: A Chinese dataset and lexicon for online sexism detection. Online Soc Network Med2022; 27: 100182.
58.
ÇÇöltekin. A corpus of Turkish offensive language on social media. In: Proceedings of the Twelfth language resources and evaluation conference. 2020, pp.6174–6184.
59.
AssenmacherDNiemannMMüllerK, et al.RP-Mod & RP-Crowd: Moderator- and crowd-annotated German news comment datasets. In: NeurIPS datasets and benchmarks track (Round 2). 2021.
60.
MoonJChoWILeeJ. BEEP! Korean corpus of online news comments for toxic speech detection. In: Proceedings of the Eighth international workshop on natural language processing for social media. 2020, pp.25–31.
61.
SarkerJSultanaSWilsonSR, et al.ToxiSpanSE: An explainable toxicity detection in code review comments. In: 2023 ACM/IEEE international symposium on empirical software engineering and measurement (ESEM), 2023, pp.1–12. IEEE.
62.
ZampieriMMalmasiSNakovP, et al.Predicting the type and target of offensive posts in social media. In: Proceedings of the 2019 conference of the north american chapter of the association for computational linguistics. 2019, pp.1415–1420.
63.
TouvronHMartinLStoneK, et al.LLaMA 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023.
64.
GuoDYangDZhangH, et al.Deepseek-r1: Incentivizing reasoning capability in LLMs via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025.
65.
HuEJShenYWallisP, et al.LoRA: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021.
66.
DaiWYuTLiuZ, et al.Kungfupanda at SemEval-2020 Task 12: BERT-based multi-task learning for offensive language detection. In: Proceedings of the fourteenth workshop on semantic evaluation. 2020, pp.2060–2066.
