Identification of Correct Triples on Open Information Extraction

Abstract

Knowledge Graph (KG) is a key component of many information systems. Due to the drastic growth of the text data, Open Information Extraction (Open IE) becomes one of the techniques to build Knowledge Graph. Nevertheless, OpenIE usually generates many noisy triples. Populating a knowledge graph with such triples degrades its quality. In this study, we introduced novel features to identify correct triples, extracted by the OpenIE system. Experimental results showed that our features achieved better results compared to the baseline features.