DABS is a single-pass framework that builds a depth-ordered substrate from one Transformer encoding and performs lightweight aspect-conditioned readout, cutting computation by up to 60% on multi-aspect ATSA benchmarks while matching prior accuracy.
CONTRASTE : Supervised Contrastive Pre-training With Aspect-based Prompts For Aspect Sentiment Triplet Extraction
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Single-Pass, Depth-Selective Reading for Multi-Aspect Sentiment Analysis
DABS is a single-pass framework that builds a depth-ordered substrate from one Transformer encoding and performs lightweight aspect-conditioned readout, cutting computation by up to 60% on multi-aspect ATSA benchmarks while matching prior accuracy.