---
title: "DUTIR at SemEval-2023 Task 10 — paper by Yu et al. (2023)"
description: "The paper presents the DUTIR team's entry in Subtask A of Task 10 at SemEval-2023, a shared task on explainably detecting… 12 references."
authors: ["Bingjie Yu", "Zewen Bai", "Haoran Ji", "Shiyi Li", "Hao Zhang", "Hongfei Lin"]
year: 2023
venue: "Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)"
source: "https://aclanthology.org/2023.semeval-1.123/"
doi: "10.18653/v1/2023.semeval-1.123"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/yu-2023-dutir-semeval-2023-task-10/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/yu-2023-dutir-semeval-2023-task-10/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/17/) / DUTIR at SemEval-2023 Task 10: Semi-supervised Learning for Sexism Detection in English

# DUTIR at SemEval-2023 Task 10: Semi-supervised Learning for Sexism Detection in English

**Authors:** Bingjie Yu, Zewen Bai, Haoran Ji, Shiyi Li, [Hao Zhang](https://md.research.elamur.ai/free/en/authors/hao-zhang/) and [Hongfei Lin](https://md.research.elamur.ai/free/en/authors/hongfei-lin/) · **Year:** 2023 · **Venue:** [Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2023](https://md.research.elamur.ai/free/en/venues/semeval/2023/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2023.semeval-1.123/) · [PDF](https://aclanthology.org/2023.semeval-1.123.pdf) · [DOI 10.18653/v1/2023.semeval-1.123](https://doi.org/10.18653/v1/2023.semeval-1.123) · [OpenAlex](https://openalex.org/W4385572359)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper presents the DUTIR team's entry in Subtask A of Task 10 at SemEval-2023, a shared task on explainably detecting sexist content online (EDOS). The task is a binary decision on whether an English post from Gab or Reddit is sexist. The organizers released about 20,000 labeled posts and roughly 2 million unlabeled ones, and the authors aimed to make good use of the unlabeled pool.

First, they continued masked-language-model pre-training of RoBERTa-base on a 200,000-post unlabeled sample plus the labeled data, which gave a domain-adapted encoder called RoBERTa A. Second, they fine-tuned it with Unsupervised Data Augmentation (UDA). This combines standard supervised loss with a consistency term that penalizes differing predictions on an unlabeled post and its EDA-noised copy. Third, they combined six models trained with different hyperparameters through majority voting.

Each stage improved results, with the largest gains on the dev set. The final system scored a test Macro-F1 of 0.8352 and ranked 41st of 84. The work shows that cheap domain pre-training and consistency training offer modest but consistent gains in a low-label classification setting.

## Key points
- Labeled data is split into 14,000 training, 2,000 dev and 4,000 test posts, with about a quarter labeled sexist. To handle this imbalance, `pos_weight` of BCEWithLogitsLoss was set to 4.
- Domain pre-training used MLM on RoBERTa-base for 700 epochs with batch size 16. The corpus was one tenth of the 2M unlabeled posts plus all labeled entries.
- The UDA loss combines cross-entropy on labeled data with KL divergence between predictions on clean and EDA-augmented unlabeled posts. The ratio α of unlabeled to labeled examples per batch ranged from 1 to 6.
- The EDA settings were synonym replacement 0.3, random insertion 0.3, random swap 0.5 and random deletion 0.3. Confidence masking used a threshold of 0.45, and sharpening used a temperature of 0.85.
- Macro-F1 scores were as follows:

| Model | Dev | Test |
|---|---|---|
| RoBERTa-base | 0.8205 | 0.8208 |
| RoBERTa A | 0.8386 | 0.8289 |
| RoBERTa A + UDA | 0.8545 | 0.8348 |
| RoBERTa A + UDA + voting | — | 0.8352 |

## Interesting ideas
- Continuing self-supervised pre-training on unlabeled text from the same source as the labeled set helped every downstream configuration at almost no annotation cost.
- Consistency training enforces that a model's output stays stable when the input is slightly perturbed, which turns unlabeled text into a training signal.
- Confidence-based masking applies the unlabeled loss only where the model is already fairly sure, so its own noisy guesses are not reinforced.
- Voting over models that differed only in hyperparameters added almost nothing, and the authors attribute this to the models being too similar.
- Checkpoints were selected on the dev set, and the gains looked larger on dev than on test (UDA +1.6 vs +0.6 points).

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (tree): gparse could get a short MLM pass on the target genre (fables, school texts, word problems) before treebank fine-tuning, to narrow the domain gap.
- **I2** (learn): unlabeled sentences could be perturbed with syntax-safe edits, such as same-POS synonym swaps, and the parser or frame builder penalized or flagged when its analysis changes. Random swaps from EDA would break the syntax and should be avoided here.
- **I3** (learn): silver trees and frames could be admitted only above a confidence or agreement threshold, which keeps self-training from amplifying errors.
- **I4** (learn): witness parsers should differ in architecture (perceptron versus gparse, or different training data) so that their agreement actually carries evidence.
- **I5** (learn): evaluations should report a held-out set never used for early stopping or model selection, so that dev-tuned improvements are not overstated.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Dalian University of Technology](https://md.research.elamur.ai/free/en/universities/dalian-university-of-technology/) — University, Dalian, China · 20 works · 46 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (2): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,894 works · [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works

## Cite

Cite as: Bingjie Yu, Zewen Bai, Haoran Ji, Shiyi Li, Hao Zhang and Hongfei Lin. 2023. DUTIR at SemEval-2023 Task 10: Semi-supervised Learning for Sexism Detection in English. Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{yu2023dutir,
  title = {{DUTIR at SemEval-2023 Task 10: Semi-supervised Learning for Sexism Detection in English}},
  author = {Bingjie Yu, Zewen Bai, Haoran Ji, Shiyi Li, Hao Zhang and Hongfei Lin},
  booktitle = {Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)},
  year = {2023},
  doi = {10.18653/v1/2023.semeval-1.123},
  url = {https://doi.org/10.18653/v1/2023.semeval-1.123}
}
```

- [← previous Discourse Analysis of Argumentative Essays of English Learners Based on CEFR Level](https://md.research.elamur.ai/free/en/papers/hanel-2023-discourse-analysis-argumentative-essays-english/)
- [next → Endalew Assefa, Descriptive grammar of Ezha, a Gurage language of Ethiopia (Ethio-Semitic)](https://md.research.elamur.ai/free/en/papers/bulakh-2023-endalew-assefa-descriptive-grammar-ezha/)
- [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/17/)

## References

12 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Harika Abburi, Pulkit Parikh, Niyati Chhaya, and Vasudeva Varma. 2020. Semi-supervised multi-task learning for multi-label fine-grained sexism classification. In Proceedings of the 28th International Conference on Computational Linguistics, page 5810–5820, Barcelona, Spain (Online). International Committee on Computational Linguistics. [find](https://scholar.google.com/scholar?q=Harika+Abburi%2C+Pulkit+Parikh%2C+Niyati+Chhaya%2C+and+Vasudeva+Varma%2E+2020%2E+Semi%2Dsupervised+multi%2Dtask+learning+for+multi%2Dlabel+fine%2Dgrained+sexism+classification%2E+In+Proceedings+of+the+28th+International+)
2. Maria Anzovino, Elisabetta Fersini, and Paolo Rosso. 2018. Automatic identification and classification of misogynistic language on twitter. In Natural Language Processing and Information Systems, page 57–64, Cham. Springer International Publishing. [find](https://scholar.google.com/scholar?q=Maria+Anzovino%2C+Elisabetta+Fersini%2C+and+Paolo+Rosso%2E+2018%2E+Automatic+identification+and+classification+of+misogynistic+language+on+twitter%2E+In+Natural+Language+Processing+and+Information+Systems%2C+page)
3. Pinkesh Badjatiya, Shashank Gupta, Manish Gupta, and Vasudeva Varma. 2017. Deep learning for hate speech detection in tweets. In Proceedings of the 26th International Conference on World Wide Web Companion, WWW ’17 Companion, page 759–760, Republic and Canton of Geneva, CHE. International World Wide Web Conferences Steering Committee. [find](https://scholar.google.com/scholar?q=Pinkesh+Badjatiya%2C+Shashank+Gupta%2C+Manish+Gupta%2C+and+Vasudeva+Varma%2E+2017%2E+Deep+learning+for+hate+speech+detection+in+tweets%2E+In+Proceedings+of+the+26th+International+Conference+on+World+Wide+Web+Comp)
4. Janet K. Swim, Robyn Mallett, and Charles Stangor. 2004. Understanding subtle sexism: Detection and use of sexist language. Sex Roles, 51(3):117–128. [find](https://scholar.google.com/scholar?q=Janet+K%2E+Swim%2C+Robyn+Mallett%2C+and+Charles+Stangor%2E+2004%2E+Understanding+subtle+sexism%3A+Detection+and+use+of+sexist+language%2E+Sex+Roles%2C+51%283%29%3A117%E2%80%93128%2E)
5. Jason Wei and Kai Zou. 2019. [Eda: Easy data augmentation techniques for boosting performance on text classification tasks](https://md.research.elamur.ai/free/en/papers/wei-2019-eda-easy-data-augmentation-techniques/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), page 6382–6388, Hong Kong, China. Association for Computational Linguistics.
6. Qizhe Xie, Zihang Dai, Eduard Hovy, Minh-Thang Luong, and Quoc V. Le. 2020. [Unsupervised data augmentation for consistency training](https://md.research.elamur.ai/free/en/papers/xie-2019-unsupervised-data-augmentation-consistency-training/). In Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS’20, Red Hook, NY, USA. Curran Associates Inc.
7. Patricia Chiril, Véronique Moriceau, Farah Benamara, Alda Mari, Gloria Origgi, and Marlène Coulomb-Gully. 2020. An annotated corpus for sexism detection in french tweets. In Proceedings of the Twelfth Language Resources and Evaluation Conference, page 1397–1403, Marseille, France. European Language Resources Association. [find](https://scholar.google.com/scholar?q=Patricia+Chiril%2C+V%C3%A9ronique+Moriceau%2C+Farah+Benamara%2C+Alda+Mari%2C+Gloria+Origgi%2C+and+Marl%C3%A8ne+Coulomb%2DGully%2E+2020%2E+An+annotated+corpus+for+sexism+detection+in+french+tweets%2E+In+Proceedings+of+the+Twelfth)
8. Thomas Davidson, Dana Warmsley, Michael Macy, and Ingmar Weber. 2017. [Automated hate speech detection and the problem of offensive language](https://md.research.elamur.ai/free/en/papers/davidson-2017-automated-hate-speech-detection-problem/). Proceedings of the International AAAI Conference on Web and Social Media, 11(1):512–515.
9. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. [Bert: Pre-training of deep bidirectional transformers for language understanding](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/). In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), page 4171–4186, Minneapolis, Minnesota. Association for Computational Linguistics.
10. Luis Espinosa Anke, Thierry Declerck, Dagmar Gromann, Ziqi Zhang, Lei Luo, Dagmar Gromann, Luis Espinosa Anke, and Thierry Declerck. 2019. Hate speech detection: A solved problem? the challenging case of long tail on twitter. Semant. Web, 10(5):925–945. [find](https://scholar.google.com/scholar?q=Luis+Espinosa+Anke%2C+Thierry+Declerck%2C+Dagmar+Gromann%2C+Ziqi+Zhang%2C+Lei+Luo%2C+Dagmar+Gromann%2C+Luis+Espinosa+Anke%2C+and+Thierry+Declerck%2E+2019%2E+Hate+speech+detection%3A+A+solved+problem%3F+the+challenging+case)
11. Hannah Rose Kirk, Wenjie Yin, Bertie Vidgen, and Paul Röttger. 2023. [SemEval-2023 Task 10: Explainable Detection of Online Sexism](https://md.research.elamur.ai/free/en/papers/kirk-2023-semeval-2023-task-10-explainable/). In Proceedings of the 17th International Workshop on Semantic Evaluation, Toronto, Canada. Association for Computational Linguistics.
12. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. [Roberta: A robustly optimized bert pretraining approach](https://md.research.elamur.ai/free/en/papers/liu-2019-roberta-robustly-optimized-bert-pretraining/). ArXiv, abs/1907.11692. [↗](https://arxiv.org/abs/1907.11692)

*Updated 2 November 2026*