Elamur Research⌕
☰
home / papers / 2026 / McMaster NLP at SemEval-2026 Task 2: A Lightweight Multi-Feature System for Predicting Emotional Valence and Arousal over Time

McMaster NLP at SemEval-2026 Task 2: A Lightweight Multi-Feature System for Predicting Emotional Valence and Arousal over Time

Paper · 2026 · Hongyi Zhang, Daniel Hu and Allison Lahnala

Summary

SemEval-2026 Task 2, Subtask 1 asks systems to estimate continuous valence and arousal scores for texts that the same people wrote repeatedly over time. The data was collected in 2021–2024 from U.S. service-industry workers. It mixes long ecological essays with short lists of feeling words, and each text is paired with the author's self-reported affect.

The McMaster NLP team built a deliberately simple regressor that represents each entry with four feature groups:

  • all-MiniLM-L6-v2 sentence embeddings
  • LIWC category counts
  • similarity scores to LLM-generated archetype sentences for low, neutral and high affect, embedded with mxbai-embed-large-v1 and inspired by PHQ-9 item scoring
  • a learned vector for each user

An MLP maps the combined features to both targets at once. The authors report that this design is a strong, interpretable baseline, and valence is predicted more reliably than arousal. Ablations show that lexical-semantic cues drive valence, while arousal barely changes when any single feature group is removed. Narrative essays and syntactically complex texts produce larger errors, especially for arousal. The study shows that cheap, transparent pipelines remain viable for tracking affect over time.

Key points

  • Variance from user differences: differences between users account for roughly 36.6% of the variation in valence and 17.5% of the variation in arousal. This motivates a trainable 32-dimensional user embedding. Users not seen in training get a zero vector, so their predictions rely only on text features.
  • Regressor: the MLP has hidden layers of 512, 256 and 128 units, with ReLU and dropout. It is trained with MSE loss and Adam at a learning rate of 10⁻⁴. The training data was split 80/20 at the instance level for validation and early stopping.
  • LIWC preprocessing: the submitted system used raw LIWC counts. After submission, the authors added length normalization, a log-token-count feature and StandardScaler.
  • Known limitations: each text is predicted independently, with no temporal modeling of user trajectories. The archetype sentences were written by an LLM rather than by experts. The labels are subjective self-reports.
  • Code: available at github.com/McMasterNLP/SemEval-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

🎓McMaster UniversityHamilton · Canada6 works · 656 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 worksSocial Psychology · 330 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 1,035 works

Cite

Cite as Hongyi Zhang, Daniel Hu and Allison Lahnala. 2026. McMaster NLP at SemEval-2026 Task 2: A Lightweight Multi-Feature System for Predicting Emotional Valence and Arousal over Time. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.98

@inproceedings{zhang2026mcmaster,
  title = {{McMaster NLP at SemEval-2026 Task 2: A Lightweight Multi-Feature System for Predicting Emotional Valence and Arousal over Time}},
  author = {Hongyi Zhang, Daniel Hu and Allison Lahnala},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.98},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.98}
}

References

26 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Prashant Bharadwaj, Mallesh M. Pai, and Agne Suziede-lyte. 2017. Mental health stigma. Economics Letters, 159:57–60. find
  2. Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna Wallach. 2020. Language (technology) is power: A critical survey of “bias” in NLP. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5454–5476. Association for Computational Linguistics.
  3. Ryan L. Boyd and H. Andrew Schwartz. 2021. Natural language analysis and the psychology of verbal behavior: The past, present, and future states of the field. Journal of Language and Social Psychology, 40(1):21–41. find
  4. Margaret M. Bradley and Peter J. Lang. 1994. Measuring emotion: The self-assessment manikin and the semantic differential. Journal of Behavior Therapy and Experimental Psychiatry, 25(1):49–59. find
  5. Trevor Cohen, Benjamin Blatter, and Vimla L. Patel. 2008. Simulating expert clinical comprehension: Adapting latent semantic analysis to accurately extract clinical concepts from psychiatric narrative. Journal of Biomedical Informatics, 41(6):1070–1087. find
  6. Kevin Doherty, Andreas Balaskas, and Gavin Doherty. 2020. The design of ecological momentary assessment technologies. Interacting with Computers, 32(3):257–278. find
  7. J. C. Eichstaedt, R. J. Smith, R. M. Merchant, L. H. Ungar, P. Crutchley, D. Preoţiuc-Pietro, D. A. Asch, and H. A. Schwartz. 2018. Facebook language predicts depression in medical records. Proceedings of the National Academy of Sciences of the United States of America, 115(44):11203–11208. find
  8. Zhijun Guo, Alvina Lai, Johan H. Thygesen, Joseph Farrington, Thomas Keen, and Kezhi Li. 2024. Large language models for mental health applications: Systematic review. JMIR Mental Health, 11:e57400. find
  9. Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings.
  10. Sean Lee, Aamir Shakir, Darius Koenig, and Julius Lipp. 2024. Open source strikes bread - new fluffy embeddings model. find
  11. Xianming Li and Jing Li. 2023. Angle-optimized text embeddings. arXiv preprint arXiv:2309.12871.
  12. Siddharth Mangalik, Johannes C. Eichstaedt, Salvatore Giorgi, Jihu Mun, Farhan Ahmed, Gilvir Gill, Adithya V. Ganesan, Shashanka Subrahmanya, Nikita Soni, Sean A. P. Clouston, and H. Andrew Schwartz. 2024. Robust language-based mental health assessments in time and space through social media. npj Di… find
  13. Marwan Omar, Soohyeon Choi, DaeHun Nyang, and David Mohaisen. 2022. Robust natural language processing: Recent advances, challenges, and future directions. IEEE Access, 10:86038–86056. find
  14. James W. Pennebaker. 2018. Expressive writing in psychological science. Perspectives on Psychological Science, 13(2):226–229. find
  15. James W Pennebaker, Roger J Booth, Ryan L Boyd, and Martha E Francis. 2015. Linguistic inquiry and word count: Liwc2015. find
  16. Jonathan Posner, James A. Russell, and Bradley S. Peterson. 2005. The circumplex model of affect: an integrative approach to affective neuroscience, cognitive development, and psychopathology. Development and Psychopathology, 17(3):715–734. find
  17. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992,…
  18. James A. Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178. find
  19. Nikita Soni, H. Andrew Schwartz, Ryan L. Boyd, Phi Long Bui, Syeda Mahwish, August Håkan Nilsson, Adithya V Ganesan, Lyle Ungar, Niranjan Balasubramanian, and Saif M. Mohammad. 2026. SemEval-2026 task 2: Predicting variation in emotional valence and arousal over time from ecological essays. In Proce…
  20. Yla R. Tausczik and James W. Pennebaker. 2010. The psychological meaning of words: Liwc and computerized text analysis methods. Journal of Language and Social Psychology, 29(1):24–54. find
  21. G. K. Verma and U. S. Tiwary. 2017. Affect representation and recognition in 3d continuous valence– arousal–dominance space. Multimedia Tools and Applications, 76:2159–2183. find
  22. Huy Vu, Huy Anh Nguyen, Adithya V. Ganesan, Swanie Juhng, Oscar N. E. Kjell, Joao Sedoc, Margaret L. Kern, Ryan L. Boyd, Lyle Ungar, H. Andrew Schwartz, and Johannes C. Eichstaedt. 2024. Psychadapter: Adapting llm transformers to reflect traits, personality and mental health. find
  23. B. R. Whitehead and C. S. Bergeman. 2013. Self-reported health bias: the role of daily affective valence and arousal. Psychology & Health, 28(7):784– 799. find
  24. T. Zhang, A. M. Schoene, S. Ji, and et al. 2022. Natural language processing applied to mental illness detection: a narrative review. npj Digital Medicine, 5:46. find
  25. Deyu Zhou, Jianan Wang, Linhai Zhang, and Yulan He. 2021. Implicit sentiment analysis with event-centered text representation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6884–6893, Online and Punta Cana, Dominican Republic. Association for Comput… find
  26. Scale 5 4 3 2 1 Valence Rating Pleasant Pleased Neutral Unsatisfied Unpleasant Arousal Rating Excited Wide-awake Neutral Dull Calm *Table 1: Five-point scale used for self-reported valence and arousal ratings. Valence ranges from unpleasant (1) to pleasant (5), while arousal ranges from calm (1) to… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-mcmaster-nlp-semeval-2026-task/