Pixel Phantoms at SemEval-2026 Task 3: Language-Specific Transformer Regression for Dimensional Aspect-Based Sentiment Analysis
Year: 2026
Type: conference-paper
Abstract: Aspect-Based Sentiment Analysis (ABSA) has traditionally relied on discrete polarity labels that fail to capture the continuous nature of human emotion.SemEval-2026 Task 3, Dimensional Aspect-Based Sentiment Analysis (DimABSA), addresses this limitation by requiring continuous Valence and Arousal prediction on a 1-9 scale for specific aspect terms across 15 language-domain combinations.The Pixel Phantoms system adopts a languageaware strategy, selecting dedicated pre-trained transformer models for each language and falling back to xlm-roberta-base for lowerresource cases such as Tatar and Ukrainian.All models share a common regression architecture with dual pooling, aspect-prompted input formatting, and a composite MSE + MAE loss.We participated in both Track A and Track B, with our strongest result in Japanese Hotel (rank 13/21, RMSE 0.7297) and competitive performance in Chinese restaurant (RMSE 0.9823 vs. Baseline Kimi-K2 Thinking 1.8959).Overall, language-specific encoders provide clear gains over generic multilingual baselines for dimensional sentiment regression.
Authors: Jithu Morrison S, Abisha Rose S
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166884929