Deep Learning-Based Knowledge Graph Generation for COVID-19

Many attempts have been made to construct new domain-specific knowledge graphs using the existing base of various domains. However, traditional “dictionary-based” or “supervised” graph building methods rely on predefined human-annotated resources entities and their relationships. The cost creating is high in terms both time effort. This means that relying will not allow rapid adaptability describing when information added updated very frequently, such as with recent coronavirus disease-19 (COVID-19) pandemic situation. Therefore, this study, we propose an Open Information Extraction (OpenIE) system based unsupervised learning without a pre-built dataset. proposed method obtains from vast amount text documents about COVID-19 rather than general add graph. First, constructed entity dictionary, then scraped large dataset related COVID-19. Next, perspective language model by fine-tuning bidirectional encoder representations transformer (BERT) pre-trained model. Finally, defined COVID-19-specific extracting connecting words between BERT self-attention weight sentences. Experimental results demonstrated Co-BERT outperforms original mask prediction accuracy metric for evaluation translation explicit ordering (METEOR) score.