Dear colleagues,

We are delighted to announce *SemEval-2026 Task 3: Dimensional Aspect-Based Sentiment Analysis on Customer Reviews and Stance Datasets*.

*Aspect-Based Sentiment Analysis (ABSA)* is a widely used technique for analyzing people’s opinions and sentiments at the aspect level. However, current ABSA research predominantly adopts a coarse-grained, categorical sentiment representation (e.g., positive, negative, or neutral). This approach stands in contrast to long-established theories in psychology and affective science, where sentiment is represented along fine-grained, real-valued dimensions of valence (ranging from negative to positive) and arousal (from sluggish to excited). This valence-arousal (VA) representation has inspired the rise of dimensional sentiment analysis as an emerging research paradigm, enabling more nuanced distinctions in emotional expression and supporting a broader range of applications.

To bridge this gap, we propose *Dimensional ABSA (DimABSA)*, a shared task that integrates dimensional sentiment analysis into the traditional ABSA framework. Furthermore, there is a conceptual similarity between stance detection and ABSA when the stance target is treated as an aspect. Building on this, we introduce *Dimensional Stance Analysis (DimStance)*, a Stance-as-DimABSA task that reformulates stance detection under the ABSA schema in the VA space. This new formulation extends ABSA beyond consumer reviews to public-issue discourse (e.g., social, political, energy, climate) and also generalizes stance analysis from categorical labels to continuous VA scores.

——————— *Languages* ——————— *We provide data in 9 languages*, including: German (deu), English (eng), Hausa (hau), Japan (jpn), Russian (rus), Swahili (swa), Tatar (tat), Ukrainian (ukr), and Chinese (zho)

——————— *Domains* ——————— *A total of 6 application domains*, including: Restaurant, Laptop, Hotel, Finance, Environmental Protection, and Politics

——————— *Subtasks* ——————— *Track A – Dimensional Aspect-Based Sentiment Analysis (DimABSA)*: Predict real-valued valence–arousal (VA) scores for aspects and extract their associated information from text. Its subtasks include: - *Subtask 1: DimASR *– Dimensional Aspect Sentiment Regression - *Subtask 2: DimASTE* – Dimensional Aspect Sentiment Triplet Extraction - *Subtask 3: DimASQP* – Dimensional Aspect Sentiment Quad Prediction

*Track B – Dimensional Stance Analysis (DimStance)*: A Stance-as-DimABSA task, where the target in stance detection is treated as an aspect. Its subtasks include: - Subtask 1: DimASR for stance analysis

——————— *Evaluation* ——————— For both tracks, RMSE is used for Subtask 1, and a new metric (continuous F1) for Subtasks 2 & 3.

——————— *Participation* ——————— *Website* (checkout details): https://github.com/DimABSA/DimABSA2026

*Codabench* (register and submit results) - Track A: https://www.codabench.org/competitions/10918/ - Track B: https://www.codabench.org/competitions/11139/

*Discord* (community and discussion) https://discord.gg/xWXDWtkMzu

*Google Group* (official updates): https://groups.google.com/g/dimabsa-participants

——————— *Important Dates * ——————— - Sample Data Ready: 15 July 2025 - Training Data Ready: 30 September 2025 - Evaluation Start: 12 January 2026 - Evaluation End: 30 January 2026 - System Description Paper Due: February 2026 - Notification to Authors: March 2026 - Camera Ready Due: April 2026 - SemEval Workshop 2026: co-located with ACL 2026 (San Diego, CA, USA)

We warmly invite the community to participate in this exciting shared task and contribute to advancing NLP research.

Best regards, SemEval-2026 Task 3 Organizers