Review
Open access
Aug 2026
Parameter-Efficient Adaptation and Benchmarking of Large Vision-Language Models for Multimodal Aspect-Based Sentiment Analysis
The resulting performance degradation under shuffled-image ablation indicates that reliable aspect-level sentiment prediction depends on preserving cross-modal semantic correspondence, and demonstrates the effectiveness of parameter-efficient LVLM adaptation for MABSA.
Ismail Ifakir, E. Nfaoui, Abderrahim Zannou
· Symmetry · 0 citations