Zero-Shot Learning via Contrastive Learning on Dual Knowledge Graphs
Graph Convolutional Networks (GCNs), which can integrate both explicit knowledge and implicit knowledge together, have shown effectively for zero-shot learning problems. Previous GCN-based methods generally leverage a single category (relationship) knowledge graph for zero-shot learning. However, in practical scenarios, multiple types of relationships among categories are usually available which can be represented as multiple knowledge graphs. To this end, we propose a novel dual knowledge graph contrastive learning framework to perform zero-shot learning. The proposed model fully exploits multiple relationships among different categories for zero-shot learning by employing graph convolutional representation and contrastive learning techniques. The main benefit of the proposed contrastive learning module is that it can effectively encourage the consistency of the category representations from different knowledge graphs while enhancing the discriminability of the generated category classifiers. We perform extensive experiments on several benchmark datasets and the experimental results show the superior performance of our approach.