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 11 languages*, including: German (deu), English (eng), Hausa (hau), Japan (jpn), Kinyarwanda (kin), Russian (rus), Swahili (swa), Tatar (tat), Twi（twi), Ukrainian (ukr), and Chinese (zho)
——————— *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: To be announced soon.
*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: 10 January 2026* - Evaluation End 31 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