Determining Relation Semantics by Mapping Relation Phrases to Knowledge Base

Open Information Extraction (Open IE) recognizes domain-independent relational triples from natural language texts but does not recognize the exact semantics of the triples. Nevertheless, relations in existing knowledge bases, like manually edited attributes in Wikipedia Info boxes, have provided the semantic meaning of open relations. Therefore, our goal is to bridge the gap between open relation phrases extracted by an Open IE system and a pre-defined attributes in knowledge base. This task is important for Information Extraction. In this paper, we proposed a novel approach to map relation phrases to attributes in knowledge base. We matched relational triples extracted from Wikipedia articles with attribute triples generated from Wikipedia Info boxes. Finally, our system expanded 509 Wikipedia attributes with a total of 2,527 phrases. Experiments show that the proposed approach can achieve an average precision of 0.88±0.02 over all mapped relation pairs.