GRAPH: A Domain Ontology-driven Semantic Graph Auto Extraction System

This paper presents sGRAPH – a domain ontology-driven semantic graph auto extraction system used to discover knowledge from text publications in traditional Chinese medicine. The traditional Chinese medicine language system (TCMLs), composed of an ontology schema and a knowledge base containing 153,692 words and 304,114 relations, is used as the domain ontology. The sGRAPH comprises two components: a user interface that interacts with users and the domain ontology-based semantic graph extraction algorithm. This algorithm is divided into five steps: text processing, semantic graph extraction, graph identification, keyword-based semantic graph search and the selectable enrichment to the knowledge base. When the knowledge base of TCMLs is used, the domain-specific words are extracted from sentences more accurately; and the hierarchical structure of the ontology can also be used to help identify the extracted graphs. The algorithm not only can extract relations between words that have already been annotated by relations in the knowledge base but also can predict the relations between words that have never been annotated by relations. The sGRAPH was developed and evaluated by extracting semantic graphs from 2000 publications which predicted 6778 relations that have never been found.