Adaptive contextual memory graph transformer with domain-adaptive knowledge graph for aspect-based sentiment analysis

Aspect-Based Sentiment Analysis (ABSA) faces critical challenges, including contextual variability, implicit sentiment inference, fluctuating sentiment intensity, and domain adaptation limitations. Traditional approaches struggle with accurately capturing domain-specific sentiment shifts, handling i...

Full description

Saved in:
Bibliographic Details
Published in:Expert systems with applications Vol. 278; p. 127300
Main Authors: Dubey, Gaurav, Dubey, Anil Kumar, Kaur, Kamaljit, Raj, Gaurav, Kumar, Parveen
Format: Journal Article
Language:English
Published: Elsevier Ltd 10.06.2025
Subjects:
Aspect knowledge representation
Multi-relational learning
Sentiment adaptation
Aspect knowledge representation
Sentiment adaptation
Multi-relational learning
ISSN:0957-4174
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Aspect-Based Sentiment Analysis (ABSA) faces critical challenges, including contextual variability, implicit sentiment inference, fluctuating sentiment intensity, and domain adaptation limitations. Traditional approaches struggle with accurately capturing domain-specific sentiment shifts, handling implicit aspects, and integrating multi-relational dependencies. To address these challenges, this study proposes the Adaptive Contextual Memory Graph Transformer (ACM-GT-ABSA), a novel framework that integrates dynamic knowledge graph construction, real-time contextual adaptation, and hierarchical sentiment representation. The Domain-Adaptive Knowledge Graph (DAKG) models domain-specific sentiment transitions and temporal variations, while the contextual memory embedding layer aligns historical sentiment trends with real-time data. A hierarchical graph encoder extracts multi-granular sentiment features, and implicit aspect detection employs contrastive learning and Semantic Role Labeling (SRL) to infer latent aspects. Additionally, hypergraph attention mechanisms model dependencies between primary and secondary intents for enhanced sentiment interpretation. Experimental evaluations on benchmark datasets (restaurant-15, restaurant-16, laptop-14 and twitter) demonstrate the model’s superior accuracy (up to 96.32%) and F1-score (88.36%), outperforming existing ABSA frameworks.
ISSN:0957-4174
DOI:10.1016/j.eswa.2025.127300