Span labeling with sentiment-aware GCN and multi-channel contrastive learning for aspect sentiment triplet extraction
Multimedia Systems · 2026-06 · 32 (4)
DOI: 10.1007/s00530-026-02364-0 ↗Kurz gefasst
Fachartikel in der Zeitschrift Multimedia Systems, online erschienen am 06.05.2026. Verfasst von You Li und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 44 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Open Project Program of Guangxi Key Laboratory of Digital Infrastructure, Innovation Project of GUET Graduate Education.
Abstract (Originalsprache)
Aspect Sentiment Triplet Extraction (ASTE) aims to identify aspect terms, opinion terms, and their sentiment dependencies from unstructured reviews. Despite recent progress, existing approaches face two fundamental limitations: (1) Vulnerability to Misaligned Syntax: Traditional syntax-based models often utilize dependency trees in a passive manner, making them vulnerable to noise propagation from inaccurate or sentiment-irrelevant syntactic relations; (2) Span boundary Ambiguity: In complex contexts involving multi-word expressions, the semantic boundaries of aspects and opinions are often blurred, leading to indistinguishable feature representations. To tackle these challenges, we propose SGMCL, a unified framework for sentiment information integration that synergizes a Sentiment-aware Graph Convolution Network (SGCN) with Multi-channel Contrastive Learning (MCL). Unlike methods that treat structure and semantics separately, SGMCL establishes a synergistic loop: the SGCN module incorporates explicit sentiment knowledge to dynamically re-weight dependency edges, thereby reshaping the information flow to align with sentiment semantics. Subsequently, the MCL mechanism operates on this refined structure, projecting aspect, opinion, and sentiment features into distinct subspaces to sharpen the discrimination of span boundaries, thereby enhancing the precision of span labeling. Experimental results on four benchmark datasets demonstrate that SGMCL achieves superior precision and competitive overall performance, particularly showing significant robustness in scenarios involving long-range dependencies and multi-word spans, proving the effectiveness of joint structural denoising and span boundary modeling.
Förderung
- National Natural Science Foundation of China
- Open Project Program of Guangxi Key Laboratory of Digital Infrastructure
- Innovation Project of GUET Graduate Education
- Guangxi Key Research and Development Program
Zitieren
APA
Li, Y., Chen, S., Peng, X., & Lin, Y. (2026). Span labeling with sentiment-aware GCN and multi-channel contrastive learning for aspect sentiment triplet extraction. Multimedia Systems, 32(4). https://doi.org/10.1007/s00530-026-02364-0
Vancouver
Li Y, Chen S, Peng X, Lin Y. Span labeling with sentiment-aware GCN and multi-channel contrastive learning for aspect sentiment triplet extraction. Multimedia Systems. 2026;32(4). doi:10.1007/s00530-026-02364-0
BibTeX
@article{Li2026,
author = {Li, You and Chen, Siyuan and Peng, Xinyong and Lin, Yuming},
title = {Span labeling with sentiment-aware GCN and multi-channel contrastive learning for aspect sentiment triplet extraction},
journal = {Multimedia Systems},
year = {2026},
volume = {32},
number = {4},
doi = {10.1007/s00530-026-02364-0}
}
EUVIMED