
Motivation & Challenge: Irrelevant context misleads aspect-level sentiment predictions. We introduced diffusion-based denoising that removes lexicographic, bag-of-words, and syntactic noise while preserving useful context.
Results & Venues: Achieved state-of-the-art results on five ABSA benchmark datasets (LAP14, REST14/15/16, MAMS).
Published in Findings of NAACL 2024.
Link: https://aclanthology.org/2024.findings-naacl.194/
1 like · 0 views · Sep 15, 2026
Motivation & Challenge: Irrelevant context misleads aspect-level sentiment predictions.