Multimedia Systems Computer Networks and Communications
CD-SIPO: a cross-dimensional sentiment analysis framework with semantic interaction and path optimization
Abstract
In Aspect-Based Sentiment Analysis (ABSA), accurately identifying sentiment polarity remains challenging, particularly for sentences with complex structures. Existing methods usually model the relationships between aspects and contextual words through dependency trees, but they may fail to capture deeper semantic relations, especially in sentences containing nested clauses or implicit sentiment expressions. To address this limitation, we propose CD-SIPO, a structure-aware framework that incorporates Abstract Meaning Representation (AMR) to provide richer semantic relations than traditional dependency structures. We further design a Semantic Structure Path Optimization Strategy (SPS), which aggregates sentiment-related semantic paths through a path aggregator and employs a Grouped Coordinate Interaction (GCI) Module to distinguish informative paths from noisy ones and enhance direction-aware semantic interactions. In addition, we introduce a Relation-Aware Feature Enhancement Attention (RFEA) Module, which models global dependencies across channel and positional dimensions to strengthen aspect-related sentiment representations and improve the fusion of contextual and structural features. Through the joint modeling of contextual semantics and optimized semantic structures, CD-SIPO captures fine-grained aspect–sentiment relationships more effectively. Experimental results on several widely used ABSA benchmarks show that the proposed framework consistently outperforms competitive baseline models in both Accuracy and Macro-F1. These results demonstrate the effectiveness of semantic path optimization and cross-dimensional feature enhancement for aspect-based sentiment classification.