Elamur Research⌕
☰
home / papers / 2026 / DreamerNLplus: Interpretable Modeling of Mental Health Dynamics from Social Media Timelines using Hybrid Rule-Based and RAG Methods

DreamerNLplus: Interpretable Modeling of Mental Health Dynamics from Social Media Timelines using Hybrid Rule-Based and RAG Methods

Paper · 2026 · Maryia Zhyrko, Daisy Lal, Erik van Mulligen et al.

Summary

The paper describes DreamerNLplus, a system submitted to every subtask of the CLPsych 2026 shared task on tracking mental health changes in anonymized social media timelines. The aim is interpretable, privacy-conscious modeling, so all models run locally.

  • Task 1 (ABCD state prediction): Prompt2Predict-DeBERTa uses Ollama to expand label evidence, fine-tunes DeBERTa to classify subelements, and passes one-hot predictions to a Random Forest that estimates presence scores. A rule-based n-gram matcher served as a comparison.
  • Task 2 (moments of change): a few-shot Llama 3.1 8B looks at each post plus up to five earlier posts. It flags Switch and Escalation events and adds a short justification.
  • Task 3 (summaries and signatures): a deterministic template pipeline (MINDTRACE-SUMMARY) was compared with few-shot open LLMs. A two-stage RAG approach mined recurrent change signatures.

The best results came in Task 3: 2nd place in 3.1, and 1st (Improvement) and 3rd (Deterioration) in 3.2. Task 1 placed mid-table. The authors also highlight metric conflicts, which point to the need for unified evaluation in clinical NLP.

Key points

  • Task 1 results: ranked 22nd in Task 1.1 (macro F1 0.254) and 20th in Task 1.2 (RMSE 1.285). Across teams, subelement F1 and presence RMSE were moderately negatively correlated (r = −0.486, p = 0.00354).
  • Task 2 baseline: XGBoost with TF-IDF, all-mpnet-base-v2 embedding differences and products, timeline position, and 14 linguistic features. Rare positive examples were up-weighted.
  • Task 3.1 approaches: the fixed five-part MINDTRACE-SUMMARY template was compared with few-shot Gemma 2 9B, Gemma 4 E4B and Llama 3.1 8B. The template achieved high CS but low CT.
  • Task 3.2 method: RAG-LLM Signature Mining first extracts patterns within batches, then synthesizes them across batches. It outputs a 90-word signature per direction with 5–10 supporting example sequences, and scored top in Specificity for Improvement.
  • Resources: code and prompts are available at github.com/4dpicture/CLPsych2026. The work was funded through the EU Horizon Europe 4D PICTURE project.

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

🎓Lancaster UniversityLancaster · United Kingdom35 works · 1,615 citations🎓Leiden UniversityLeiden · Netherlands32 works · 1,074 citations🤝Leiden University Medical CenterLeiden · Netherlands4 works · 2 citations

Also on the byline: Erasmus MC.

 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 Maryia Zhyrko, Daisy Lal, Erik van Mulligen and Lifeng Han. 2026. DreamerNLplus: Interpretable Modeling of Mental Health Dynamics from Social Media Timelines using Hybrid Rule-Based and RAG Methods. 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-1.36

@inproceedings{zhyrko2026dreamernlplus,
  title = {{DreamerNLplus: Interpretable Modeling of Mental Health Dynamics from Social Media Timelines using Hybrid Rule-Based and RAG Methods}},
  author = {Maryia Zhyrko, Daisy Lal, Erik {van Mulligen} and Lifeng Han},
  booktitle = {Proceedings of the 11th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.clpsych-1.36},
  url = {https://doi.org/10.18653/v1/2026.clpsych-1.36}
}

References

16 references, 0 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. Samuel Belkadi, Nicolo Micheletti, Lifeng Han, Warren Del-Pinto, and Goran Nenadic. 2025. Lt3: Generating medication prescriptions with conditional transformer. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), pages 205–218. find
  5. Liliia Bogdanova, Shiran Sun, Lifeng Han, Natalia Amat Lefort, and Flor Miriam Plaza-del Arco. 2026. Flans at semeval-2026 task 7: Rag with open-sourced smaller llms for everyday knowledge across diverse languages and cultures. arXiv preprint arXiv:2603.01910. find
  6. Callum Chan, Sunveer Khunkhun, Diana Inkpen, and Juan Antonio Lossio-Ventura. 2025. Prompt engineering for capturing dynamic mental health self states from social media posts. In Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2025), pages 256–267, Albu… find
  7. Lifeng Han, David Lindevelt, Sander Puts, Erik van Mulligen, and Suzan Verberne. 2026. Dutch metaphor extraction from cancer patients’ interviews and forum data using llms and human in the loop. CL4Health WS at LREC2026, Palma, Spain. find
  8. Arshia Kermani, Veronica Perez-Rosas, and Vangelis Metsis. 2025. A systematic evaluation of LLM strategies for mental health text analysis: Fine-tuning vs. prompt engineering vs. RAG. In Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2025), pages 172–1… find
  9. Aziliz Le Glaz, Yannis Haralambous, Deok-Hee Kim-Dufor, Philippe Lenca, Romain Billot, Taylor C Ryan, Jonathan Marsh, Jordan Devylder, Michel Walter, Sofian Berrouiguet, and 1 others. 2021. Machine learning and natural language processing in mental health: systematic review. Journal of medical Inter… find
  10. Matteo Malgaroli, Thomas D Hull, James M Zech, and Tim Althoff. 2023. Natural language processing for mental health interventions: a systematic review and research framework. Translational Psychiatry, 13(1):309. find
  11. Libo Ren, Yee Man Ng, and Lifeng Han. 2025. Malei at multiclinsum: Summarisation of clinical documents using perspective-aware iterative self-prompting with llms. find
  12. Pablo Romero, Libo Ren, Lifeng Han, and Goran Nenadic. 2025. The manchester bees at peranssumm 2025: Iterative self-prompting with claude and o1 for perspective-aware healthcare answer summarisation. In Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health), pages 340… 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. Talia Tseriotou, Jenny Chim, Ayal Klein, Aya Shamir, Guy Dvir, Iqra Ali, Cian Kennedy, Guneet Singh Kohli, Anthony Hills, Ayah Zirikly, Dana Atzil-Slonim, and Maria Liakata. 2025. Overview of the CLPsych 2025 shared task: Capturing mental health dynamics from social media timelines. In Proceedings o… find
  15. A CLPsych 2026 Shared Tasks CLPsych2026 is affiliated with The Workshop on Computational Linguistics and Clinical Psychology, a workshop series founded in 2014. find
  16. CLPsych 2026 will be held at ACL in San Diego, July 4th, 2026. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhyrko-2026-dreamernlplus-interpretable-modeling-mental-health/