Year: 2026
Type: article
Abstract: Multimodal Aspect-Based Sentiment Analysis (MABSA) aims to predict the sentiment polarity toward a given aspect term from social-media image–text pairs. Prior approaches often under-exploit hierarchical visual cues and do not explicitly leverage syntactic constraints to bind aspect terms with their sentiment-bearing opinion expressions, which weakens aspect-specific cross-modal alignment under noisy visual content. To this end, we propose a Multi-Granularity Gated Fusion Network (MGGFN) that couples syntax-aware evidence modeling with multi-granularity visual–text representation learning. MGGFN constructs a weighted graph by combining probabilistic dependency weights with semantic similarities between visually enhanced textual nodes. Scene-level captions and face-level descriptions provide global visual-semantic context. An adaptive gating mechanism then integrates global context and local evidence for aspect-level sentiment prediction. Experiments on Twitter-2015 and Twitter-2017 show that MGGFN achieves mean macro-F1 scores of 73.2 and 69.6, respectively, across three runs. Among the compared methods, MGGFN demonstrates competitive performance on both datasets, supporting its effectiveness for aspect-level multimodal sentiment prediction. The source code is publicly available at: https://github.com/BlackHand12138/MGGFN-MABSA.git .
Source: Scientific Reports
Institution: North China Institute of Aerospace Engineering
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7234804828
