Elamur Research⌕
☰
home / papers / 2026 / Hierarchical Multi-Stage Modeling of Adaptive and Maladaptive Self-States in Social Media Timelines

Hierarchical Multi-Stage Modeling of Adaptive and Maladaptive Self-States in Social Media Timelines

Paper · 2026 · Abir Naskar and Mike Conway

Summary

The paper presents a system for Subtask 1 of the CLPsych 2026 Shared Task. The task is to identify adaptive and maladaptive self-states in Reddit user timelines using the ABCD/MIND framework. Systems must detect ABCD subelements, back them with evidence spans, and rate each state's centrality on a 1–5 scale. The labelled data is small: 30 training timelines (373 posts), plus 10 unlabelled test timelines (92 posts).

The authors test two routes. The first is a two-stage encoder built on BERT or RoBERTa:

  • Stage one decides whether a post contains a self-state and regresses its presence score.
  • Stage two starts from the stage-one encoder and reads the post paired with an element identifier. It blocks subelement labels that are invalid for that element and tags evidence tokens with BIO labels.

The second route chains four separately LoRA-tuned stages of Qwen3-4B-Instruct.

Qwen3 scores best, but RoBERTa comes very close at much lower compute cost. This suggests that a structured, CPU-friendly design can compete with larger generative models in clinical NLP.

Key points

  • Reported scores (subelement Macro-F1 / RMSE): Qwen3 0.333 / 1.06, RoBERTa 0.331 / 1.146, BERT 0.313 / 1.255, and the LLaMA-3.1-8B-Instruct one-shot baseline 0.247 / 1.424. BERT was the official submission (team NoviceTrio). RoBERTa and Qwen3 were evaluated after the competition. The paper describes RoBERTa's result as an 8.3% Macro-F1 gain over the baseline.
  • Training objectives:
    • Stage 1 combines cross-entropy and MSE, with the MSE term weighted by λ = 0.5.
    • Stage 2 combines masked cross-entropy over the label set (16 adaptive, 16 maladaptive, plus none) with BCE on BIO evidence tags.
    • Class imbalance is handled with a weighted random sampler rather than a weighted loss.
  • Training setup:
    • Encoders: learning rate 2e-5, batch size 16, sequence length 256/300.
    • Qwen3: tuned via Unsloth with 4-bit LoRA, learning rate 2e-4, maximum length 2048.
  • Ablations used a single 85/15 split. Freezing the Stage 2 encoder caused the largest drop. Removing the presence head, the weighted sampling, or the element masking also hurt results.
  • The system trails the top-ranked team by about 0.11. Encoder models produce far lower carbon emissions than Qwen3. Code is available at github.com/AbeerNaskar/clpsych26-sharedtask1.

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

🎓The University of MelbourneMelbourne · Australia100 works · 22,618 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksPsychology · 1,552 worksExperimental and Cognitive Psychology · 870 worksSocial Psychology · 306 works

Cite

Cite as Abir Naskar and Mike Conway. 2026. Hierarchical Multi-Stage Modeling of Adaptive and Maladaptive Self-States in Social Media Timelines. Proceedings of the 11th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.clpsych-1.37

@inproceedings{naskar2026hierarchical,
  title = {{Hierarchical Multi-Stage Modeling of Adaptive and Maladaptive Self-States in Social Media Timelines}},
  author = {Abir Naskar and Mike Conway},
  booktitle = {Proceedings of the 11th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.clpsych-1.37},
  url = {https://doi.org/10.18653/v1/2026.clpsych-1.37}
}

References

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

  1. Iqra Ali, Talia Tseriotou, Guy Dvir, Callum Chan, Yuxiang Zhou, Juan Antonio Lossio-Ventura, Ayal Klein, Aya Shamir, Dan Sayda, Anthony Hills, Aya Zirikly, Diana Inkpen, Dana Atzil-Slonim, and Maria Liakata. 2026. Overview of the clpsych 2026 shared task: Capturing and characterizing mental health c… find
  2. Dana Atzil-Slonim. 2025. Multimodal intrapersonal and interpersonal dynamics (mind): A transtheoretical coding manual. find
  3. Dana Atzil-Slonim. 2026. Leveraging theoretical and technological innovations to study the mechanisms that underlie therapeutic change in psychotherapy. In Louis G. Castonguay, Dana Atzil-Slonim, Michael Barkham, and Wolfgang Lutz, editors, Practice-Based Evidence in the Psychological Therapies: Tow… find
  4. Adrian Benton, Glen Coppersmith, and Mark Dredze. 2017. Ethical research protocols for social media health research. In Proceedings of the First ACL Workshop on Ethics in Natural Language Processing, pages 94–102, Valencia, Spain. Association for Computational Linguistics.
  5. 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,…
  6. Giulia Ferraro and Luca Benedetti. 2025. Hierarchical multi-task learning for fine-grained and coarse text classification. Frontiers in Interdisciplinary Applied Science, 2(2):184–190. find
  7. Edward J Hu, yelong shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  8. 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. CoRR, abs/1907.11692.
  9. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F Christiano, Jan Leike, and Ryan Lowe. 2022. Tra… — Training language models to follow instructions with human feedback
  10. Sebastian Ruder. 2017. An overview of multi-task learning in deep neural networks. CoRR, abs/1706.05098.
  11. Dana Atzil Slonim. 2024. Self-other dynamics (sod): A transtheoretical coding manual. find
  12. Bing Tian, Yong Zhang, Jin Wang, and Chunxiao Xing. 2019. Hierarchical inter-attention network for document classification with multi-task learning. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19, pages 3569–3575. International Joint Conferenc… find
  13. Adam Tsakalidis, Jenny Chim, Iman Munire Bilal, Ayah Zirikly, Dana Atzil-Slonim, Federico Nanni, Philip Resnik, Manas Gaur, Kaushik Roy, Becky Inkster, Jeff Leintz, and Maria Liakata. 2022. Overview of the CLPsych 2022 shared task: Capturing moments of change in longitudinal user posts. In Proceedin… find
  14. Demonstrations, pages 38–45, Online. Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/naskar-2026-hierarchical-multi-stage-modeling-adaptive/