Elamur Research⌕
☰
home / papers / 2026 / VISHC at PsyDefDetect: Mitigating Data Scarcity in Psychological Defense Classification with Context-Aware Synthetic Augmentation

VISHC at PsyDefDetect: Mitigating Data Scarcity in Psychological Defense Classification with Context-Aware Synthetic Augmentation

Paper · 2026 · Hoang-Thuy-Duong Vu, Quoc-Cuong Pham and Huy-Hieu Pham

Summary

The paper describes the VISHC team's system for the PsyDefDetect shared task (BioNLP@ACL 2026). The task is to classify psychological defense mechanisms in seeker utterances from the PsyDefConv dataset, which is derived from ESConv. Annotated data are scarce and the label distribution is highly skewed. To address this, the authors augment minority classes with synthetic utterances from Llama-3-8B-Instruct. Each prompt is anchored to a stressor identified in the dialogue and includes detailed class definitions taken from the Defense Mechanisms Rating Scales (DMRS). Generated batches are kept only if a secondary classifier agrees with their labels.

For classification, a hybrid late-fusion model combines three inputs: MentalRoBERTa embeddings, lightweight linguistic heuristics, and a 30-dimensional DMRS defense profile. The profile is built by scoring 150 defense indicators with an NLI model.

On the blind test set, the system clearly outperforms the DMRS Co-Pilot baseline in accuracy and macro-F1. The authors attribute most of the gain to richer definitions in the prompts. However, a strong bias toward the majority class shows that augmentation alone does not solve class imbalance.

Key points

  • Leaderboard result: the official submission (×8 augmentation) reached 58.26% accuracy and 24.62% macro-F1, compared with 18.01% and 8.63% for DMRS Co-Pilot. It ranked 13th of 21 teams.
  • Effect of augmentation scale: ×2 augmentation gave the best macro-F1 (27.99%). Larger scales performed worse, with macro-F1 dropping to 22.38% at ×10, which the authors link to accumulated synthetic noise.
  • Quality control: minority classes were capped at 500 samples, Class 7 (High-Adaptive) was not augmented, and only batches with Cohen's κ ≥ 0.60 were retained.
  • Fusion architecture: a 768-dim MentalRoBERTa vector is concatenated with two 32-dim MLP outputs, one for heuristic features and one for DMRS features (832 dims in total), and fed into a classifier over 9 labels.
  • Label 7 "sink" effect: Label 7 accounts for 243 of the 472 test instances and attracts errors from other classes; 77% of Label 5 (Neurotic) cases are predicted as Label 7.
  • Remaining weaknesses: only Labels 0 and 7 exceed an F1 of 0.70. Labels 0 and 8 have no DMRS defense items, so the DMRS features describe them poorly.

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

🎓VinUniversityHanoi · Vietnam11 works · 56 citations

 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 worksPsychology · 1,552 worksSocial Psychology · 306 works

Cite

Cite as Hoang-Thuy-Duong Vu, Quoc-Cuong Pham and Huy-Hieu Pham. 2026. VISHC at PsyDefDetect: Mitigating Data Scarcity in Psychological Defense Classification with Context-Aware Synthetic Augmentation. Proceedings of the BioNLP 2026 (Shared Tasks).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.bionlp-2.12

@inproceedings{vu2026vishc,
  title = {{VISHC at PsyDefDetect: Mitigating Data Scarcity in Psychological Defense Classification with Context-Aware Synthetic Augmentation}},
  author = {Hoang-Thuy-Duong Vu, Quoc-Cuong Pham and Huy-Hieu Pham},
  booktitle = {Proceedings of the BioNLP 2026 (Shared Tasks)},
  year = {2026},
  doi = {10.18653/v1/2026.bionlp-2.12},
  url = {https://doi.org/10.18653/v1/2026.bionlp-2.12}
}

References

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

  1. detection. In 2017 IEEE International Conference on Computer Vision (ICCV), pages 2999–3007. find
  2. Ateret Anaby-Tavor, Boaz Carmeli, Esther Goldbraich, Amir Kantor, George Kour, Segev Shlomov, Naama Tepper, and Naama Zwerdling. 2020. Do not have enough data? Deep learning to the rescue! Proceedings of the AAAI Conference on Artificial Intelligence, 34(05):7383–7390. find
  3. Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang, and Minlie Huang. 2021. Towards emotional support dialog systems. In Proceedings of ACL-IJCNLP 2021, pages 3469–3483. Association for Computational Linguistics.
  4. Michael Bond, Susan Gardner, J. Christian, and J. Sigal. 1983. An empirical study of self-rated defense style. Archives of General Psychiatry, 40:333–338. find
  5. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Conference on Learning Representations. find
  6. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… — Language Models are Few-Shot Learners
  7. Hongbin Na, Yining Hua, Zimu Wang, Tao Shen, Beibei Yu, Lilin Wang, Wei Wang, John Torous, and Ling Chen. 2025. A survey of large language models in psychotherapy: Current landscape and future directions. In Findings of the Association for Computational Linguistics: ACL 2025, pages 7362–7376, Vienna… find
  8. Hongbin Na, Zimu Wang, Zhaoming Chen, Yining Hua, Rena Gao, Kailai Yang, Ling Chen, Wei Wang, Shaoxiong Ji, John Torous, and Sophia Ananiadou. 2026a. Overview of the psydefdetect shared task at bionlp 2026: Detecting levels of psychological defense mechanisms in supportive conversations. In Proceedi…
  9. Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20:37 – 46. find
  10. Phebe Cramer. 1987. The development of defense mechanisms. Journal of Personality, 55(4):597–614. find
  11. Mariagrazia Di Giuseppe and J. Christopher Perry. 2021. The hierarchy of defense mechanisms: Assessing defensive functioning with the defense mechanisms rating scales Q-sort. Frontiers in Psychology, 12:718440. find
  12. Hongbin Na, Zimu Wang, Zhaoming Chen, Peilin Zhou, Yining Hua, Grace Ziqi Zhou, Haiyang Zhang, Tao Shen, Wei Wang, John Torous, Shaoxiong Ji, and Ling Chen. 2026b. You never know a person, you only know their defenses: Detecting levels of psychological defense mechanisms in supportive conversations.…
  13. Shaoxiong Ji, Tianlin Zhang, Luna Ansari, Jie Fu, Prayag Tiwari, and Erik Cambria. 2022. Mental-BERT: Publicly available pretrained language models for mental healthcare. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 7184–7190, Marseille, France. European Langu… find
  14. George E. Vaillant. 1994. Ego mechanisms of defense and personality psychopathology. Journal of Abnormal Psychology, 103(1):44–50. find
  15. Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. 2023. Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12):1–38.
  16. Jason Wei and Kai Zou. 2019. EDA: Easy data augmentation techniques for boosting performance on text classification tasks. 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…
  17. Douwe Kiela, Suvrat Bhooshan, Hamed Firooz, Ethan Perez, and Davide Testuggine. 2020. Supervised multimodal bitransformers for classifying images and text. Preprint, arXiv:1909.02950. find
  18. Yaoming Zhu, Sidi Lu, Lei Zheng, Jiaxian Guo, Weinan Zhang, Jun Wang, and Yong Yu. 2018. Texygen: A benchmarking platform for text generation models. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, SIGIR ’18, page 1097–1100, New York, NY, USA. Assoc…
  19. Varun Kumar, Ashutosh Choudhary, and Eunah Cho. 2020. Data augmentation using pre-trained transformer models. In Proceedings of the 2nd Workshop on Life-long Learning for Spoken Language Systems, pages 18–26, Suzhou, China. Association for Computational Linguistics.
  20. Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vu-2026-vishc-psydefdetect-mitigating-data-scarcity/