Biomedical temporal knowledge graph reasoning via contrastive adversarial learning

In the biomedical domain, Temporal knowledge graph (TKG) has emerged as a valuable method to capture dynamic evolutional relationships and predict future associations between entities. However, entity representations that evolve over time are more difficult to learn compared to static representations, especially for those entities that lack historical information or emerge recently. In this paper, we propose a novel Contrastive Adversarial Learning (CAL) framework for TKG reasoning in biomedical domain. We utilize adversarial training approach to increase entity samples, in order to enhance robustness and generalization ability of the model. Contrastive learning is also adopted to more accurately distinguish different entities that change over time. Specifically, we first construct three biomedical TKGs mainly based on the Comparative Toxicogenomics Database (CTD). Then, an evolution encoder is adopted to obtain temporal evolutional embeddings of entities. Next, we generate adversarial samples for entity embeddings by adding a perturbation to each of them. Finally, the obtained adversarial samples are used for contrastive learning to capture the temporal evolution of entity relationships. Experiments on three biomedical datasets show that our CAL framework achieves state-of-the-art performance.