Multimodal self supervised adversarial graph attention neural network

Multimodal self supervised adversarial graph attention neural network

연구 분야: Artificial Intelligence

학회: 2024 5th International Conference on Computer Vision, Image and Deep Learning (CVIDL)

The emergence of short video platforms has spurred the fusion of personalized recommendation systems with visual, textual, and auditory modalities. However, the sparsity of data limits current research, impacting model representation. To address this, we introduce Multimodal Self-Supervised Generative Adversarial Graph Attention Learning (MSGAT). It leverages self-supervised adversarial learning for data augmentation and gate attention mechanisms with cross-modal contrastive learning to capture complex user interactions. Our experiments on real-world datasets validate our method’s superiority, achieving over a 6% increase in accuracy, NDCG, and Recall compared to the best baseline model.