Elamur Research⌕
☰
home / papers / 2026 / CS_Metro at PsyDefDetect: Detecting Psychological Defense Mechanisms in Mental Health Dialogues with Summarization-Enhanced Transformer Ensembles

CS_Metro at PsyDefDetect: Detecting Psychological Defense Mechanisms in Mental Health Dialogues with Summarization-Enhanced Transformer Ensembles

Paper · 2026 · Oarisa Rebayet, Radiul Walee, Symom Hossain Shohan et al.

Summary

The paper describes the CS_Metro team's system for the PsyDefDetect shared task at BioNLP 2026. The task is to assign one of nine DMRS-based defense levels to a help-seeker utterance in a therapeutic dialogue, using the PSYDEFCONV corpus. This is hard because defenses depend on context and the label distribution is heavily skewed toward the High Adaptive class.

The authors built a three-stage pipeline:

  1. Qwen3-4B-Instruct condenses each dialogue into a short summary of at most 100 words, which serves as global context.
  2. Mental-health-pretrained encoders (Mental-BERT, Mental-RoBERTa, Mental-XLNet) are fine-tuned on the data. Training uses undersampling of the dominant class and inverse-frequency class weights.
  3. A rule-based ensemble chooses, for each class, the prediction of the model that performed most reliably on that class during validation.

They compared this pipeline with fine-tuned LLMs (Qwen3-4B, Qwen3-1.7B, Mistral-7B) and with few-shot prompting. On the released gold test labels, the ensemble reached 34.69% macro F1, the best of all configurations tested. The results suggest that compact domain-specific encoders can beat generic LLMs on niche clinical classification.

Key points

  • Data: PSYDEFCONV has 200 dialogues and 2,336 labeled utterances. High Adaptive (Label 7) makes up about 52% of the data, and its training count was reduced from 968 to 550 samples.
  • Main result: The ensemble scored 34.69% macro F1, 4.69 points above the baseline. Fine-tuned Qwen3-4B reached 30.04%, and the best few-shot LLM reached only 19.48%.
  • Effect of summarization varied by model: Mental-XLNet gained 7.48 F1 points with summaries. Mental-BERT improved slightly, and Mental-RoBERTa got worse.
  • Official leaderboards: The system scored 23.46% macro F1 on Leaderboard 1 (labels 1–8), ranking 15th of 21 teams. On Leaderboard 2 (labels 0–8) it scored 30.04%, placing 14th.
  • Errors and code: High Adaptive absorbed many errors, with minority classes such as Disavowal, Neurotic and Minor Image Distorting often misclassified as it. The code is available on GitHub (OarisaR/CS_Metro-at-PsyDefDetect-BioNLP-2026).

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

🎓Chittagong University of Engineering & TechnologyChittagong · Bangladesh25 works · 29 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

🧭 Topics 2Research subjects and languages the work is about.Dialogue systems · 624 worksSummarization · 232 works

Cite

Cite as Oarisa Rebayet, Radiul Walee, Symom Hossain Shohan, Kawsar Ahmed and Mohammed Moshiul Hoque. 2026. CS_Metro at PsyDefDetect: Detecting Psychological Defense Mechanisms in Mental Health Dialogues with Summarization-Enhanced Transformer Ensembles. 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-2.26

@inproceedings{rebayet2026cs,
  title = {{CS\_Metro at PsyDefDetect: Detecting Psychological Defense Mechanisms in Mental Health Dialogues with Summarization-Enhanced Transformer Ensembles}},
  author = {Oarisa Rebayet, Radiul Walee, Symom Hossain Shohan, Kawsar Ahmed and Mohammed Moshiul Hoque},
  booktitle = {Proceedings of the BioNLP 2026 (Shared Tasks)},
  year = {2026},
  doi = {10.18653/v1/2026.bionlp-2.26},
  url = {https://doi.org/10.18653/v1/2026.bionlp-2.26}
}

References

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

  1. Xin Bai, Guanyi Chen, Tingting He, Rui Fan, Chenlian Zhou, and Yu Liu. 2026. Emotional supporters often use multiple strategies in a single turn. Neurocomputing, 693:133897. find
  2. Stevie Chancellor and Munmun De Choudhury. 2020. Methods in predictive techniques for mental health status on social media: a critical review. npj Digital Medicine, 3(1):43. find
  3. 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. PMID: 34366995. find
  4. Sigmund Freud. 1936. Inhibitions, symptoms and anxiety. The Psychoanalytic Quarterly, 5:1–28. Original work published 1936. find
  5. 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 LREC. find
  6. Shaoxiong Ji, Tianlin Zhang, Kailai Yang, Sophia Ananiadou, Erik Cambria, and Jörg Tiedemann. 2023. Domain-specific continued pretraining of language models for capturing long context in mental health. arXiv preprint arXiv:2304.10447. — [arXiv:2304.10447](https://arxiv.org/abs/2304.10447) find
  7. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix,… find
  8. 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 the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Lan…
  9. Navid Madani and Rohini Srihari. 2025. Steering conversational large language models for long emotional support conversations. In Proceedings of the Third Workshop on Social Influence in Conversations (SICon 2025), pages 109–123, Vienna, Austria. Association for Computational Linguistics. find
  10. 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
  11. 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…
  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. J. Christopher Perry, Marianne E. Kardos, and Christopher J. Pagano. 1993. The study of defenses in psychotherapy using the defense mechanism rating scales (dmrs). In The Concept of Defense Mechanisms in Contemporary Psychology: Theoretical, Research, and Clinical Perspectives, pages 122–132. find
  14. John Perry and Melissa Henry. 2004. Studying defense mechanisms in psychotherapy using the defense mechanism rating scales. Advances in Psychology, 136. find
  15. Qwen Team. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388.
  16. George E. Vaillant. 1992. Ego Mechanisms of Defense: A Guide for Clinicians and Researchers. American Psychiatric Press. Original work published June 1, 1992. find
  17. Tenggan Zhang, Xinjie Zhang, Jinming Zhao, Li Zhou, and Qin Jin. 2024. ESCoT: Towards interpretable emotional support dialogue systems. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13395–13412, Bangkok, Thailand. Associatio… find
  18. Chujie Zheng, Sahand Sabour, Jiaxin Wen, Zheng Zhang, and Minlie Huang. 2023a. AugESC: Dialogue augmentation with large language models for emotional support conversation. In Findings of the Association for Computational Linguistics: ACL 2023, pages 1552–1568, Toronto, Canada. Association for Comput… find
  19. System Prompt You are a clinical psychologist specializing in psychodynamic defense mechanisms. Classify the psychological defense mechanism in the given therapy exchange. Defense levels (choose exactly one): • 0 – No Defense: No clear defense mechanism observed. • 1 – Action: Acting out, passive ag… find
  20. Zhonghua Zheng, Lizi Liao, Yang Deng, and Liqiang Nie. 2023b. Building emotional support chatbots in the era of llms. ArXiv, abs/2308.11584. — [arXiv:2308.11584](https://arxiv.org/abs/2308.11584) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rebayet-2026-cs-metro-psydefdetect-detecting-psychological/