Elamur Research⌕
☰
home / papers / 2026 / CCNU at SemEval-2026 Task 10: Conspiracy Marker Extraction and Detection via Multi-task Learning and LLM-based Data Augmentation

CCNU at SemEval-2026 Task 10: Conspiracy Marker Extraction and Detection via Multi-task Learning and LLM-based Data Augmentation

Paper · 2026 · Zijun Wang and Guanyi Chen

Summary

This paper describes CCNU's entry to SemEval-2026 Task 10 (PsyCoMark), which studies conspiracy thinking in Reddit comments. The task has two parts. Subtask 1 extracts spans for five narrative roles (Actor, Action, Effect, Evidence, Victim), and Subtask 2 is binary classification of whether a comment is conspiratorial.

For extraction, the authors combine all roles into one BIO label set. A single shared transformer encoder with one token-classification head learns all roles together, which lets it exploit how roles co-occur. Training data was expanded with Qwen-generated paraphrases. A five-model ensemble mixed BERT, RoBERTa and DeBERTa variants trained with and without augmentation. For detection, they used BERT and DistilBERT classifiers with heavier Qwen augmentation.

The extraction system placed 3rd on the official test set, and its gains came mainly from ensemble diversity. Detection results were competitive, but ensembling did not help there, and neither did adding predicted marker features.

The main takeaway is that the best optimization strategy depends on task granularity. Sparse span labeling benefits most from model diversity, while sentence-level classification depends more on representation quality and data variety.

Key points

  • Subtask 1 (span F1, IoU ≥ 0.5):
    • The 5-model ensemble reached 0.27 on dev and 0.25 on test, ranking 3rd.
    • The best single multi-task model, DeBERTa with augmentation, scored 0.23 on dev and 0.20 on test.
    • Single-task baselines scored 0.18–0.21 on dev.
  • Qwen paraphrase augmentation: about 2× the training data for extraction (kept moderate to preserve span alignment) and about 3× for detection (with more structural rephrasing).
  • Subtask 2 (Macro F1):
    • BERT with augmentation was best, at 0.80 on dev and 0.73 on test.
    • Augmentation gave small but consistent gains for both BERT and DistilBERT.
    • Ensembles reached only 0.70 and 0.72 on test, below the best single model.
  • Cross-task transfer: adding extracted marker indicators to the detection classifiers lowered dev Macro F1, from 0.78 to 0.72 for BERT and from 0.74 to 0.66 for RoBERTa. The authors attribute this to errors propagating from the extraction step.
  • Model size: larger encoders (DeBERTa, RoBERTa) did not reliably beat BERT or DistilBERT on detection.

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

Also on the byline: Central China Normal University.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

Cite

Cite as Zijun Wang and Guanyi Chen. 2026. CCNU at SemEval-2026 Task 10: Conspiracy Marker Extraction and Detection via Multi-task Learning and LLM-based Data Augmentation. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.58

@inproceedings{wang2026ccnu,
  title = {{CCNU at SemEval-2026 Task 10: Conspiracy Marker Extraction and Detection via Multi-task Learning and LLM-based Data Augmentation}},
  author = {Zijun Wang and Guanyi Chen},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.58},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.58}
}

References

8 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. 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,…
  2. Karen Douglas, Robbie Sutton, and Aleksandra Cichocka. 2017. The psychology of conspiracy theories. Current Directions in Psychological Science, 26. find
  3. Karen M. Douglas and Robbie M. Sutton. 2023. What are conspiracy theories? a definitional approach to their correlates, consequences, and communication. Annual Review of Psychology, 74:271–298. find
  4. Alex Fong, Jon Roozenbeek, Daniel Goldwert, Steve Rathje, and Sander van der Linden. 2021. The language of conspiracy: A psychological analysis of speech used by conspiracy theorists and their followers on twitter. Group Processes & Intergroup Relations, 24(4):606–623. find
  5. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021. Deberta: Decoding-enhanced bert with disentangled attention. Preprint, arXiv:2006.03654.
  6. 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. Preprint, arXiv:1907.11692.
  7. Mattia Samory, Felix Soldner, and Veronika Batzdorfer. 2026. SemEval-2026 task 10: PsyCoMark – psycholinguistic conspiracy marker extraction and detection. In Proceedings of the 20th International Workshop on Semantic Evaluation (SemEval-2026).
  8. Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2020. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. Preprint, arXiv:1910.01108. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-ccnu-semeval-2026-task-10/