Extracting triples from Vietnamese text to create knowledge graph

Abstract

Knowledge graph (KG) plays an increasingly important role in the current technology era. It is very useful in many fields such as searching for information, supporting question answering systems and in other AI applications, etc. Besides the private Knowledge Graphs like Google's "Knowledge graph", we also have Open Knowledge graphs as DBpedia, YAGO, ... But generally, these Open Knowledge graphs contain very little data in Vietnamese. Due to this practice, our team proposed a way to create Vietnamese Knowledge graph by automatically scratching the Vietnamese text on the website as input, then using Named-entity recognition (NER) to recognize entities as nouns and combined with POS tag identifies words as verbs to extract triple in the simple sentences of the paragraph. The triple was then loaded into Neo4j to visualize the Knowledge graph.