COVID-19 has emerged as a severe global epidemic with high morbidity and mortality, while no effective drug targets it. Drug repurposing is a promising drug discovery method by minimizing the time and cost compared to new drug discovery and traditional randomized clinical trials. Big data-driven network analytics provide a novel way for drug repurposing, but currently, no such public available graph tailor-made for COVID-19. Here, we build a comprehensive COVID-19 knowledge graph by incorporating 25 data sources, including information on drugs, genes, viruses, proteins, diseases, and symptoms, which are notable predictors of clinical efficacy. By utilizing the COVID-19 knowledge graph and network-based methodologies, we can rapidly identify candidate repurposable drugs and drug combinations by discovering complex intrinsic linkages.