Elamur Research⌕
☰
home / papers / 2026 / Momentum at SemEval-2026 Task 2: LongVA-RoBERTa, a transformer-Based Longitudinal Valence and Arousal Modeling

Momentum at SemEval-2026 Task 2: LongVA-RoBERTa, a transformer-Based Longitudinal Valence and Arousal Modeling

Paper · 2026 · Supriya Nadiger, Sunil Saumya, Rahul Pujari et al.

Summary

This paper describes the Momentum team's system for SemEval-2026 Task 2. The task asks systems to estimate emotion as continuous valence and arousal scores rather than discrete categories, using longitudinal ecological essays and feeling-word responses linked to participants' self-reported EMA ratings. The team addressed two subtasks.

  • Subtask 1 predicts scores for each text. LongVA-RoBERTa uses a partially fine-tuned RoBERTa-base encoder, single-head attention pooling and a two-layer MLP regressor. It also adds a few handcrafted surface features, such as text length, punctuation counts and an essay vs. feeling-word flag.
  • Subtask 2A forecasts short-term changes in affect between consecutive entries. Each user's entries are put in chronological order and passed through the same encoder plus a BiLSTM and MLP.

The authors report that Subtask 1 beats the official baseline across composite, between-user and within-user evaluation, though the extracted text does not include the numbers. For Subtask 2A the system is only competitive. Its valence error is slightly lower than the baseline's, but its correlations are lower. The work shows that adding sequence modelling to transformer encoders helps with longitudinal emotion analysis.

Key points

  • Data: 5,285 longitudinal entries from 182 participants, collected 2021–2024 and annotated with EMA valence and arousal ratings. The test set covers both seen and unseen users.

  • Setup:

    • RoBERTa-base with the lower 8 layers frozen and the top 4 fine-tuned, inputs capped at 128 tokens.
    • Subtask 1: L1 loss, AdamW, learning rate 2×10⁻⁵.
    • Subtask 2A: MSE loss, learning rate 1×10⁻⁴, user-grouped GroupKFold validation.

  • Subtask 2A results (system vs. baseline):

    DimensionSystem rBaseline rSystem MAEBaseline MAE
    Valence.553.6151.1391.168
    Arousal.589.6700.6980.638

  • Ablation (Subtask 2A, Pearson r on validation):

    • RoBERTa alone: 0.366 for valence, 0.312 for arousal.
    • Adding user embeddings changes little.
    • Adding the BiLSTM gives clear gains.
    • The full model reaches 0.553 for valence and 0.589 for arousal.

  • Limitation: the model treats irregular EMA time gaps as uniform steps. The authors also suggest the BiLSTM may smooth out sharp arousal spikes. Encoding elapsed time is proposed as future work.

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

🎓Indian Institute of Information Technology DharwadDharwad · India4 works · 6 citations

Also on the byline: KLE Technological University.

 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 Supriya Nadiger, Sunil Saumya, Rahul Pujari, Veeresh Hiremath, Kiran Chikaraddi and Anoop Kadkol. 2026. Momentum at SemEval-2026 Task 2: LongVA-RoBERTa, a transformer-Based Longitudinal Valence and Arousal Modeling. 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-1.78

@inproceedings{nadiger2026momentum,
  title = {{Momentum at SemEval-2026 Task 2: LongVA-RoBERTa, a transformer-Based Longitudinal Valence and Arousal Modeling}},
  author = {Supriya Nadiger, Sunil Saumya, Rahul Pujari, Veeresh Hiremath, Kiran Chikaraddi and Anoop Kadkol},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.78},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.78}
}

References

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

  1. Jacob Benesty, Jingdong Chen, Yiteng Huang, and Israel Cohen. 2009. Pearson correlation coefficient. In Noise reduction in speech processing, pages 1–4. Springer. find
  2. Sven Buechel and Udo Hahn. 2017. Emobank: Studying the impact of annotation perspective and representation format on dimensional emotion analysis. In Proceedings of the 15th Conference of the European Chapter of the ACL. find
  3. Paul Ekman. 1992. An argument for basic emotions. Cognition and Emotion, 6(3-4):169–200.
  4. Adithya V Ganesan and 1 others. 2026. From word sequences to behavioral sequences: Adapting modeling and evaluation paradigms for longitudinal nlp. arXiv preprint arXiv:2601.07988. find
  5. Pranav Goel, Devang Kulshreshtha, Prayas Jain, and Kaushal Kumar Shukla. 2017. Prayas at emoint 2017: An ensemble of deep neural architectures for emotion intensity prediction in tweets. In Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analys… find
  6. Marieke Houben, Peter Kuppens, and Francis Tuerlinckx. 2015. Emotion variability and psychological well-being: A meta-analysis. Psychological Bulletin. find
  7. Maximilian Koper, Evgeny Kim, and Roman Klinger. 2017. Ims at emoint-2017: Emotion intensity prediction with affective norms, automatically extended resources and deep learning. In Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages… find
  8. Peter Kuppens, Francis Tuerlinckx, and Marc A. B. van Mechelen. 2010. Emotion dynamics and emotional inertia. Psychological Science. find
  9. Kevin LaGrandeur. 2015. Emotion, artificial intelligence, and ethics. In Beyond Artificial Intelligence: The Disappearing Human-Machine Divide, pages 97–109. Springer. find
  10. 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. arXiv preprint arXiv:1907.11692.
  11. Sreekanth Madisetty and Maunendra Sankar Desarkar. 2017. Nsemo at emoint-2017: An ensemble to predict emotion intensity in tweets. In Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 219–224. find
  12. Samarth Tripathi, Shrinivas Acharya, Ranti Sharma, Sudhanshi Mittal, and Samit Bhattacharya. 2017. Using deep and convolutional neural networks for accurate emotion classification on deap data. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 4746–4752. find
  13. Laura Troiano and 1 others. 2023. Quantifying valence and arousal in text with multilingual pre-trained transformers. arXiv preprint arXiv:2302.14021. — [arXiv:2302.14021](https://arxiv.org/abs/2302.14021) find
  14. Talia Tseriotou, Adam Tsakalidis, and 1 others. 2023. Sequential path signature networks for personalised longitudinal language modeling. In Findings of ACL. find
  15. Cort J Willmott and Kenji Matsuura. 2005. Advantages of the mean absolute error (mae) over the root mean square error (rmse) in assessing average model performance. Climate research, 30(1):79–82. find
  16. Yue Xie, Zseiyu Liang, Zhenlin Liang, Chengwei Huang, Cairong Zou, and Björn Schuller. 2019. Speech emotion classification using attention-based lstm. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 27(11):1675–1685. find
  17. Goncalo Mendes and Bruno Martins. 2023. Quantifying valence and arousal in text with multilingual pre-trained transformers. In ECIR. find
  18. Fatemehsadat Mireshghallah and 1 others. 2022. Useridentifier: Implicit user representations for personalized sentiment analysis. In NAACL. find
  19. Saif M. Mohammad. 2016. Sentiment analysis: Detecting valence, emotions, and other affectual states from text. In Emotion Measurement, pages 201–237. Elsevier. find
  20. Saif M. Mohammad and Felipe Bravo-Marquez. 2017. Wassa-2017 shared task on emotion intensity. arXiv preprint arXiv:1708.03700.
  21. Saif M. Mohammad, Felipe Bravo-Marquez, Mohammad Salameh, and Svetlana Kiritchenko. 2018. Semeval-2018 task 1: Affect in tweets. In Proceedings of the 12th International Workshop on Semantic Evaluation, pages 1–17.
  22. James A. Russell. 1980. A circumplex model of affect. Journal of Personality and Social Psychology, 39(6):1161–1178. find
  23. 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…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nadiger-2026-momentum-semeval-2026-task-2/