Researchers have developed StanceMoE, a new Mixture-of-Experts (MoE) architecture built on a fine-tuned BERT encoder for actor-level stance detection. This model integrates six expert modules designed to capture various linguistic signals, such as semantic orientation, lexical cues, and discourse shifts. A context-aware gating mechanism dynamically weights these expert contributions for adaptive routing. Experiments on the StanceNakba 2026 Subtask A dataset showed StanceMoE achieving a macro-F1 score of 94.26%, outperforming traditional baselines and other BERT-based variants. AI
IMPACT Introduces a novel MoE architecture that improves performance on stance detection tasks, potentially influencing future research in nuanced text analysis.
RANK_REASON The cluster contains a research paper detailing a new model architecture for a specific NLP task, submitted to arXiv. [lever_c_demoted from research: ic=1 ai=1.0]
- Abdullah Al Shafi
- alphaXiv
- arXiv
- BERT
- DagsHub
- Gotit.pub
- Hugging Face
- mixture of experts
- ScienceCast
- StanceMoE
- StanceNakba 2026 Subtask A
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
