Elamur Research⌕
☰
home / papers / 2026 / HOPE: Hybrid Optimized Parallel Encoding with Supervised and Unsupervised Semantic Fusion for Depression Symptom Detection

HOPE: Hybrid Optimized Parallel Encoding with Supervised and Unsupervised Semantic Fusion for Depression Symptom Detection

Paper · 2026 · Tu-Phuong Mai, Minh-Ha H. Le, Duc-Luong Tran et al.

Summary

The paper tackles fine-grained detection of depression symptoms in social media text, framed as multi-label classification and relevance ranking. Supervised models suffer from scarce, subjective labels and from temporal domain shift between training and test data. Purely unsupervised methods mishandle texts that fall between symptom boundaries. The authors propose HOPE, which runs two views in parallel:

  • Supervised view: a fine-tuned DepRoBERTa extractor with symptom-specific heads produces relevance scores.
  • Unsupervised view: a transductive clustering module groups the unlabeled test sentences with K-means. It uses paired-stream BERT embeddings, and its centroids are seeded with clinical symptom descriptions enriched by few-shot examples. Clustering then shifts these class anchors toward the target data.

A relevance filter first removes off-topic sentences, which scatter across the embedding space. A learnable MLP then merges the supervised and clustering scores. HOPE is tested on several social-media benchmarks covering BDI-II, DSM-5, PHQ-9 and emotion label sets, and the authors report that it beats earlier systems on both ranking and classification. This suggests that unlabeled target data can add useful signal for mental-health screening.

Key points

  • The architecture has three parts: a DepRoBERTa Supervised Symptom Extractor, an Unsupervised Intrinsic Semantic Clustering module, and an Optimized Hybrid Semantic Fusion step built on an MLP.
  • The setting is transductive: test data such as the roughly 17M eRisk texts is treated as an unlabeled corpus for clustering.
  • Clustering works at the sentence level, and a post's score is the maximum over its sentences. The eRisk analysis covers 21 BDI-II-style symptoms plus a non-relevant class.
  • The error analysis identifies four error types:
    • confusion between overlapping symptoms, for example Self-Criticalness vs. Self-Dislike;
    • missed implicit symptoms;
    • questionable gold annotations;
    • incomplete predictions on sentences carrying several symptoms.
  • Code is released at github.com/candleMind/hope (ACL 2026 main conference, long paper).

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

🎓VNU University of Engineering and TechnologyHanoi · Vietnam4 works · 1 citation

 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 Tu-Phuong Mai, Minh-Ha H. Le, Duc-Luong Tran, Phuong-Anh Chu, Duy-Cat Can and Hoang-Quynh Le. 2026. HOPE: Hybrid Optimized Parallel Encoding with Supervised and Unsupervised Semantic Fusion for Depression Symptom Detection. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.132

@inproceedings{mai2026hope,
  title = {{HOPE: Hybrid Optimized Parallel Encoding with Supervised and Unsupervised Semantic Fusion for Depression Symptom Detection}},
  author = {Tu-Phuong Mai, Minh-Ha H. Le, Duc-Luong Tran, Phuong-Anh Chu, Duy-Cat Can and Hoang-Quynh Le},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.132},
  url = {https://doi.org/10.18653/v1/2026.acl-long.132}
}

References

46 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ling An, Haibo Hu, Mengke Song, Lin Cheng, Shuo Ba, Zhaocong Liu, Zhuohang Yu, Zhenyu Zhang, Yi Liu, and Chichun Zhou. 2025. Unsupervised classification for circulating tumor cells. IEEE Access, 13:85669–85681. find
  2. Eliseo Bao, Anxo Pérez, and Javier Parapar. 2024. Explainable depression symptom detection in social media. Health Information Science and Systems, 12(1):47. find
  3. Eliseo Bao, Anxo Pérez, and Javier Parapar. 2025. ReDSM5: A Reddit Dataset for DSM-5 Depression Detection. In Proceedings of the 34th ACM International Conference on Information and Knowledge Management, pages 6323–6327. find
  4. AP Bascuñana and Isabel Segura Bedmar. 2024. APB-UC3M at eRisk 2024: natural language processing and deep learning for the early detection of mental disorders. Working Notes of CLEF, pages 9–12. find
  5. AT Beck, CH Ward, M Mendelson, J Mock, and J Er-baugh. 1961. An inventory for measuring depression. Archives of general psychiatry, 4(6):561–571. find
  6. Ana-Maria Bucur, Adrian Cosma, Paolo Rosso, and Liviu P Dinu. 2023. It’s just a matter of time: Detecting depression with time-enriched multimodal transformers. In European conference on information retrieval, pages 200–215, Cham. Springer, Springer Nature Switzerland. find
  7. Sofia Chaudhary, Jennifer A Hoffmann, Christian D Pulcini, Mark Zamani, Matt Hall, Kristyn N Jeffries, Rachel Myers, Joel Fein, Bonnie T Zima, Peter F Ehrlich, and 1 others. 2024. Youth suicide and preceding mental health diagnosis. JAMA network open, 7(7):e2423996–e2423996. find
  8. Sumit Dalal, Sarika Jain, and Mayank Dave. 2024. Deep Knowledge-Infusion For Explainable Depression Detection. arXiv preprint arXiv:2409.02122. — [arXiv:2409.02122](https://arxiv.org/abs/2409.02122) find
  9. Sumit Dalal, Deepa Tilwani, Manas Gaur, Sarika Jain, Valerie L Shalin, and Amit P Sheth. 2025. A Cross Attention Approach to Diagnostic Explainability Using Clinical Practice Guidelines for Depression. IEEE Journal of Biomedical and Health Informatics, 29(2):1333–1342. find
  10. David Guecha, Aaryan Potdar, and Anthony Miyaguchi. 2024. DS@GT eRisk 2024: Sentence Transformers for Social Media Risk Assessment. Working Notes of CLEF, pages 825–833. find
  11. Lin Lawrence Guo, Stephen R Pfohl, Jason Fries, Alistair EW Johnson, Jose Posada, Catherine Aftandilian, Nigam Shah, and Lillian Sung. 2022. Evaluation of domain generalization and adaptation on improving model robustness to temporal dataset shift in clinical medicine. Scientific reports, 12(1):2726… find
  12. Shrey Gupta, Anmol Agarwal, Manas Gaur, Kaushik Roy, Vignesh Narayanan, Ponnurangam Kumaraguru, and Amit Sheth. 2022. Learning to Automate Follow-up Question Generation using Process Knowledge for Depression Triage on Reddit Posts. In Proceedings of the Eighth Workshop on Computational Linguistics a… find
  13. AbdelMoniem Helmy, Radwa Nassar, and Nagy Ram-dan. 2024. Depression detection for twitter users using sentiment analysis in English and Arabic tweets. Artificial intelligence in medicine, 147:102716. find
  14. Pervaiz Iqbal Khan, Andreas Dengel, and Sheraz Ahmed. 2024. Improving Text Representation for Disease Detection from Social Media via Self-augmentation and Contrastive Learning. In International Conference on Neural Information Processing, pages 137–151. Springer. find
  15. Andriy Kosar, Guy De Pauw, and Walter Daelemans. 2022. Unsupervised text classification with neural word embeddings. Computational Linguistics in the Netherlands Journal, 12:165–181. find
  16. Kurt Kroenke, Robert L Spitzer, and Janet BW Williams. 2001. The PHQ-9: validity of a brief depression severity measure. Journal of general internal medicine, 16(9):606–613. find
  17. Siryeol Lee, Juncheol Lee, Juntae Park, Jiwoo Park, Dohoon Kim, Joohyun Lee, and Jaehoon Oh. 2024. Deep learning-based natural language processing for detecting medical symptoms and histories in emergency patient triage. The American Journal of Emergency Medicine, 77:29–38. find
  18. Zhongchen Ma and Songcan Chen. 2021. Expand globally, shrink locally: Discriminant multi-label learning with missing labels. Pattern Recognition, 111:107675. find
  19. Tu-Phuong Mai, Minh-Ha H Le, Duc-Luong Tran, Duy-Cat Can, and Hoang-Quynh Le. 2025. UET@ eRisk2025: Severity Estimation for Depression Symptoms Searching and Early Risk Detection. Working Notes of CLEF, pages 9–12. find
  20. Kirill Milintsevich, Gaël Dias, and Kairit Sirts. 2024. Evaluating Lexicon Incorporation for Depression Symptom Estimation. In Proceedings of the 6th Clinical Natural Language Processing Workshop, pages 322–328, Mexico City, Mexico. Association for Computational Linguistics. find
  21. Quang Vinh Nguyen, Dong Thanh Nguyen, Duc Duy Nguyen, Doan Khai Ta, Hai Binh Nguyen, Ji-eun Shin, Seungwon Kim, Hyung-Jeong Yang, and Soo-Hyung Kim. 2025. A Time-Aware Mental State Space for Multimodal Depression Detection on Social Media. In Proceedings of the Annual Meeting of the Cognitive Scienc… find
  22. Diogo AP Nunes and Eugénio Ribeiro. 2025. INESC-ID@ eRisk 2025: Exploring Fine-Tuned, Similarity-Based, and Prompt-Based Approaches to Depression Symptom Identification. Working Notes of CLEF, pages 1474–1485. find
  23. Javier Parapar, Patricia Martín-Rodilla, David E Losada, and Fabio Crestani. 2023. Overview of erisk 2023: Early risk prediction on the internet. In International Conference of the Cross-Language Evaluation Forum for European Languages, pages 294–315. Springer. find
  24. Javier Parapar, Patricia Martín-Rodilla, David E. Losada, and Fabio Crestani. 2024. Overview of eRisk 2024: Early Risk Prediction on the Internet. In Experimental IR Meets Multilinguality, Multimodality, and Interaction, pages 73–92, Cham. Springer Nature Switzerland. find
  25. Javier Parapar, Anxo Perez, Xi Wang, and Fabio Crestani. 2025a. eRisk 2025: contextual and conversational approaches for depression challenges. In European Conference on Information Retrieval, pages 416–424. Springer. find
  26. Javier Parapar, Anxo Perez, Xi Wang, and Fabio Crestani. 2025b. Overview of erisk 2025: Early risk prediction on the internet. In International conference of the cross-language evaluation forum for European languages, pages 242–265. Springer. find
  27. Rafał Poświata and Michał Perełkiewicz. 2022. OPI@ LT-EDI-ACL2022: Detecting signs of depression from social media text using RoBERTa pre-trained language models. In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion, pages 276–282. find
  28. Abu Bakar Siddiqur Rahman, Hoang-Thang Ta, Lotfol-lah Najjar, Azad Azadmanesh, and Ali Saffet Gönul. 2024. DepressionEmo: A novel dataset for multilabel classification of depression emotions. Journal of Affective Disorders, 366:445–458. find
  29. Tim Schopf, Daniel Braun, and Florian Matthes. 2022. Evaluating unsupervised text classification: zero-shot and similarity-based approaches. In Proceedings of the 2022 6th International Conference on Natural Language Processing and Information Retrieval, pages 6–15. find
  30. Nguyen Minh Son and DV Thin. 2025. Sonuit eRisk2025: enhanced depression detection on social media via filtering and re-ranking. Working Notes of CLEF, pages 9–12. find
  31. Matthew Squires, Xiaohui Tao, Soman Elangovan, Raj Gururajan, Xujuan Zhou, U Rajendra Acharya, and Yuefeng Li. 2023. Deep learning and machine learning in psychiatry: a survey of current progress in depression detection, diagnosis and treatment. Brain Informatics, 10(1):10. find
  32. Phi Ta, Nha Tran, Hung Nguyen, and Hien D Nguyen. 2025. Detecting signs of depression on social media: A machine learning analysis and evaluation. Sustainable Futures, page 100827. find
  33. Vankayala Tejaswini, Korra Sathya Babu, and Bibhu-datta Sahoo. 2024. Depression detection from social media text analysis using natural language processing techniques and hybrid deep learning model. ACM Transactions on Asian and Low-Resource Language Information Processing, 23(1):1–20. find
  34. Poojan Vachharajani. 2025. Transformer ensembles and llm-powered approaches for depression symptom analysis and contextualized early risk detection. Working Notes of CLEF, pages 9–12. find
  35. Vihang N Vahia. 2013. Diagnostic and statistical manual of mental disorders 5: A quick glance. Indian journal of psychiatry, 55(3):220–223. find
  36. Marc Violides, Tanatip Timtong, Krystof Bezdek, and Pavlos Andreadis. 2024. Impact of COVID-19 on linguistic expression of depression in online communities. In Proceedings of the 2024 5th International Symposium on Artificial Intelligence for Medicine Science, pages 787–793. find
  37. Ying Zhang, Xiaocan Jia, Yongli Yang, Na Sun, Shuyan Shi, and Wei Wang. 2024. Change in the global burden of depression from 1990-2019 and its prediction for 2030. Journal of psychiatric research, 178:16– 22. find
  38. Zhiling Zhang, Siyuan Chen, Mengyue Wu, and Kenny Zhu. 2022. Symptom Identification for Interpretable Detection of Multiple Mental Disorders on Social Media. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 9970–9985, Abu Dhabi, United Arab Emirates. A… find
  39. Hamad Zogan, Imran Razzak, Shoaib Jameel, and Guandong Xu. 2021. Depressionnet: learning multi-modalities with user post summarization for depression detection on social media. In proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval, pages 1… find
  40. A • Source code is available from https://github.com/candleMind/hope. • The eRisk dataset is available upon request to the eRisk organizers at https://erisk. irlab.org/eRisk2025.html. • The ReDSM5 dataset is hosted on Hugging Face https://huggingface.co/datasets/ irlab-udc/redsm5 and requires reque… ↗
  41. Qin Wang, Cees Taal, and Olga Fink. 2021. Integrating expert knowledge with domain adaptation for unsupervised fault diagnosis. IEEE Transactions on Instrumentation and Measurement, 71:1–12. find
  42. Yuxi Wang, Diana Inkpen, and Prasadith Kirinde Gamaarachchige. 2024. Explainable depression detection using large language models on social media data. In Proceedings of the 9th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2024), pages 108–126. find
  43. Yutong Wu, David Conlan, Siegfried Perez, and Anthony Nguyen. 2024. Leveraging Unlabeled Clinical Data to Boost Performance of Risk Stratification Models for Suspected Acute Coronary Syndrome. In AMIA Annual Symposium Proceedings, volume 2023, page 744. find
  44. Aqsa Younas, Shazia Riaz, Saqib Ali, Rafiullah Khan, Mohib Ullah, and Daehan Kwak. 2025. Stacked LSTM Model for Contextual Correlation Detection Among Multiple Emotions. IEEE Access, 13:117558–117570. find
  45. Tao Yu, Wei Huang, Xin Tang, and Duosi Zheng. 2025. A hybrid unsupervised machine learning model with spectral clustering and semi-supervised support vector machine for credit risk assessment. PloS one, 20(1):e0316557. find
  46. Tianlin Zhang, Kailai Yang, Hassan Alhuzali, Boyang Liu, and Sophia Ananiadou. 2023. PHQ-aware depressive symptoms identification with similarity contrastive learning on social media. Information Processing & Management, 60(5):103417. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mai-2026-hope-hybrid-optimized-parallel-encoding/