Elamur Research⌕
☰
home / papers / 2026 / The Spectrum of Sentiment: Optimistic, Pessimistic, and Neutral Voices in Online Depression Discourse

The Spectrum of Sentiment: Optimistic, Pessimistic, and Neutral Voices in Online Depression Discourse

Summary

The authors present what they describe as the first computational study of how people with diagnosed depression express optimism and pessimism when posting on social media. They fine-tuned a RoBERTa-based classifier, called RoBERTa-OPT-3Labels, on the OPT Twitter corpus. Unlike earlier work, they kept OPT's ambiguous mid-range tweets as a separate neutral class instead of dropping them. They then applied the classifier to Reddit posts from the eRisk 2021 depression dataset. In that dataset, users count as depressed or control based on their BDI-II questionnaire scores. To check how well the model transfers to Reddit, they built a small manually labelled evaluation subset. They analysed the labelled posts with LIWC, statistical significance tests and BERTopic topic modelling, and used SHAP to find the words behind misclassifications. The main finding is that the two groups post optimistic and pessimistic content in roughly the same proportions. However, the linguistic features and topics of those posts differ in meaningful ways. This suggests that depression shows up more in how people express outlook than in how often.

Key points

  • Training data: OPT (Ruan et al., 2016) contains 7,475 tweets rated by five annotators on a scale from −3 to +3 (Krippendorff's alpha 0.731). Average scores below −1 were labelled pessimistic, scores from −1 to 1 neutral, and scores above 1 optimistic.
  • Analysis data: eRisk 2021 covers 170 Reddit users: 90 in the training split with 46,502 posts and 80 in the test split with 32,237 posts. Users scoring above the BDI-II cut-off of 19 were treated as depressed.
  • Model: the base model was twitter-roberta-base-sentiment-latest, fine-tuned with a learning rate of 5e-5, 3 epochs, a maximum length of 128 tokens, batch size 8, AdamW, early stopping and a 70/15/15 data split.
  • Evaluation subset: 150 eRisk posts were split evenly across the three classes and both user groups, then labelled by three annotators following the OPT protocol.
  • Analysis tools: LIWC was used for psycholinguistic features, BERTopic for topics and SHAP for error analysis.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Discourse analysis

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3eqsahrxy75e

@inproceedings{tabusca2026spectrum,
  title = {{The Spectrum of Sentiment: Optimistic, Pessimistic, and Neutral Voices in Online Depression Discourse}},
  author = {Tabusca, Stefana Arina and Bucur, Ana-Maria and Dinu, Liviu P.},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/3eqsahrxy75e},
  url = {https://doi.org/10.63317/3eqsahrxy75e}
}

References

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

  1. A. Alshahrani, M. Ghaffari, K. Amirizirtol, and X. Liu. 2020. Identifying optimism and pessimism in twitter messages using xlnet and deep consensus. find
  2. In International Joint Conference on Neural Networks (IJCNN). find
  3. A. Alshahrani, M. Ghaffari, K. Amirizirtol, and X. Liu. 2021. Optimism/pessimism prediction of twitter messages and users using bert with soft label assignment. In International Joint Conference on Neural Networks (IJCNN). find
  4. Mostafa M. Amin, E. Cambria, Björn Schuller, and E. Cambria. 2023. Will affective computing emerge from foundation models and general artificial intelligence? a first evaluation of chatgpt. IEEE Intelligent Systems, 38:15–23. find
  5. Mario Ezra Aragon, Adrian Pastor Lopez-Monroy, Luis Carlos González-Gurrola, and Manuel Montes-y Gómez. 2021. Detecting mental disorders in social media through emotional patterns-the case of anorexia and depression. IEEE transactions on affective computing, 14(1):211–222. find
  6. Fabio Barbieri, José Camacho-Collados, Lisa Ellen Anke, and Leandro Neves. 2020. Tweeteval: Unified benchmark and comparative evaluation for tweet classification. In Findings of the Association for Computational Linguistics: EMNLP 2020. Association for Computational Linguistics. find our card
  7. Aaron T Beck, Robert A Steer, Gregory K Brown, et al. 1996. Beck depression inventory. find
  8. G. Blanco and A. Lourenço. 2022. Optimism and pessimism analysis using deep learning on covid-19 related twitter conversations. Information Processing and Management, 59(3):102918. find
  9. Ryan L Boyd, Ashwini Ashokkumar, Sarah Seraj, and James W Pennebaker. 2022. The development and psychometric properties of liwc-22. Austin, TX: University of Texas at Austin, 10. find
  10. M. R. Broome, K. E. A. Saunders, P. J. Harrison, and S. Marwaha. 2015. Mood instability: Significance, definition and measurement. British Journal of Psychiatry, 207(4):283–285. find
  11. A. Bucur, M. Zampieri, and L. P. Dinu. 2021. An exploratory analysis of the relation between offensive language and mental health. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 3600–3606. find
  12. Ana-Maria Bucur, Berta Chulvi, Adrian Cosma, and Paolo Rosso. 2024. The expression of happiness in social media of individuals reporting depression. IEEE Transactions on Affective Computing. find
  13. J. Camacho-Collados, K. Rezaee, T. Riahi, A. Ushio, D. Loureiro, D. Antypas, J. Boisson, L. E. Anke, F. Liu, and E. M. Camara. 2022. Tweetnlp: Cutting-edge natural language processing for social media. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing: System… find
  14. C. Caragea, L. P. Dinu, and B. Dumitru. 2018. Exploring optimism and pessimism in twitter using deep learning. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). find
  15. Ş. Cobeli, I. Iordache, S. Yadav, C. Caragea, L.P. Dinu, and D. Iliescu. 2022. Detecting optimism in tweets using knowledge distillation and linguistic analysis of optimism. In International Conference on Language Resources and Evaluation. find
  16. A. De Santana Correia and E. L. Colombini. 2022. Attention, please! a survey of neural attention models in deep learning. Artificial Intelligence Review, 55(8):6037–6124. find
  17. M. Eisemann. 1984. Leisure activities of depressive patients. Acta Psychiatrica Scandinavica, 69(1):45–51. find
  18. Maarten Grootendorst. 2020. KeyBERT: Minimal keyword extraction with BERT. https://doi.org/10.5281/zenodo.4461265. — [doi:10.5281/zenodo.4461265](https://doi.org/10.5281/zenodo.4461265) ↗
  19. Maarten Grootendorst. 2022. Bertopic: Neural topic modeling with a class-based tf-idf procedure. find our card
  20. Yuting Guo, Yao Ge, Yuan-Chi Yang, Mohammed Ali Al-Garadi, and Abeed Sarker. 2022. Comparison of pretraining models and strategies for health-related social media text classification. In Healthcare, volume 10, page 1478. MDPI. find
  21. U. Herwig, A. B. Brühl, T. Kaffenberger, T. Baumgärtner, H. Boeker, and L. Jäncke. 2009. Neural correlates of ’pessimistic’ attitude in depression. Psychological Medicine, 40(5):789–800. find
  22. P. Himmelstein, S. Barb, M. A. Finlayson, and K. D. Young. 2018. Linguistic analysis of the autobiographical memories of individuals with major depressive disorder. PloS one, 13(11):e0207814. find
  23. Catherine Hobbs, Petra Vozarova, Aarushi Sabharwal, Punit Shah, and Katherine Button. 2022. Is depression associated with reduced optimistic belief updating? Royal Society Open Science, 9(2):190814. find
  24. Jutta Karhu, Juha Veijola, and Mirka Hintsanen. 2024. The bidirectional relationships of optimism and pessimism with depressive symptoms in adulthood–a 15-year follow-up study from northern finland birth cohorts. Journal of Affective Disorders, 362:468–476. find
  25. Christoph W Korn, Tali Sharot, Hendrik Walter, Hauke R Heekeren, and Raymond J Dolan. 2014. Depression is related to an absence of optimistically biased belief updating about future life events. Psychological medicine, 44(3):579–592. find
  26. Kim Kronström, Hasse Karlsson, Hermann Nabi, Tuula Oksanen, Paula Salo, Noora Sjösten, Marianna Virtanen, Jaana Pentti, Mika Kivimäki, and Jussi Vahtera. 2011. Optimism and pessimism as predictors of work disability with a diagnosis of depression: a prospective cohort study of onset and recovery. Jo… find
  27. Grace Y Lim, Wilson W Tam, Yanxia Lu, Cyrus S Ho, Melvyn W Zhang, and Roger C Ho. 2018. Prevalence of depression in the community from 30 countries between 1994 and 2014. Scientific reports, 8(1):2861. find
  28. Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V. Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. OpenReview. find
  29. D. E. Losada and F. Crestani. 2016. A test collection for research on depression and language use. In Lecture notes in computer science, pages 28–39. find
  30. Scott Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. arXiv preprint arXiv:1705.07874. find our card
  31. N. Mor and J. Winquist. 2002. Self-focused attention and negative affect: A meta-analysis. Psychological Bulletin, 128(4):638–662. find
  32. Lise Solberg Nes and Suzanne C. Segerstrom. 2006. Dispositional optimism and coping: A meta-analytic review. Personality and Social Psychology Review, 10(3):235–251. find
  33. Debora Nozza and Dirk Hovy. 2023. The state of profanity obfuscation in natural language processing scientific publications. In Findings of the Association for Computational Linguistics: ACL 2023, pages 3897–3909. find
  34. Javier Parapar, Patricia Martín-Rodilla, David E Losada, and Fabio Crestani. 2021. Overview of erisk at clef 2021: Early risk prediction on the internet (extended overview). CLEF (Working Notes), 1:864–887. find
  35. Radim Rehurek and Petr Sojka. 2011. Gensim– python framework for vector space modelling. NLP Centre, Faculty of Informatics, Masaryk University, Brno, Czech Republic, 3(2). find
  36. Jonathan Rottenberg. 2005. Mood and emotion in major depression. Current Directions in Psychological Science, 14(3):167–170. find
  37. X. Ruan, S. R. Wilson, and R. Mihalcea. 2016. Finding optimists and pessimists on twitter. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). find
  38. Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108. find
  39. Michael F Scheier, Charles S Carver, and Michael W Bridges. 2001. Optimism, pessimism, and psychological well-being. Optimism & pessimism: Implications for theory, research, and practice., pages 189–216. find
  40. Andrew M Subica, J Christopher Fowler, Jon D Elhai, B Christopher Frueh, Carla Sharp, Erin L Kelly, and Jon G Allen. 2014. Factor structure and diagnostic validity of the beck depression inventory–ii with adult clinical inpatients: Comparison to a gold-standard diagnostic interview. Psychological as… find
  41. Hilary Tindle, Bea Herbeck Belnap, Patricia R Houck, Sati Mazumdar, Michael F Scheier, Karen A Matthews, Fanyin He, and Bruce L Roll-man. 2012. Optimism, response to treatment of depression, and rehospitalization after coronary artery bypass graft surgery. Psychosomatic medicine, 74(2):200–207. find
  42. Ana-Sabina Uban, Berta Chulvi, and Paolo Rosso. 2021. An emotion and cognitive based analysis of mental health disorders from social media data. Future Generation Computer Systems, 124:480– 494. find
  43. Ana Sabina Uban, Berta Chulvi, and Paolo Rosso. 2022. Multi-aspect transfer learning for detecting low resource mental disorders on social media. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 3202–3219. find
  44. Michael von Glischinski, Ruth von Brachel, and Gerrit Hirschfeld. 2019. How depressed is “depressed”? a systematic review and diagnostic meta-analysis of optimal cut points for the beck depression inventory revised (bdi-ii). Quality of Life Research, 28:1111–1118. find
  45. Dong Xu, Yi-Lun Wang, Kun-Tang Wang, Yue Wang, Xin-Ran Dong, Jie Tang, and Yuan-Lu Cui. 2021. A scientometrics analysis and visualization of depressive disorder. Current neuropharmacology, 19(6):766–786. find
  46. Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and Quoc V. Le. 2019. Xlnet: Generalized autoregressive pretraining for language understanding. arXiv preprint arXiv:1906.08237. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tabusca-2026-spectrum-sentiment-optimistic-pessimistic-neutral/