Phrase-based clause extraction for open information extraction system

Abstract

Recent development of variety and volume of information circulating in the Internet has prompted the emergence of a new paradigm in information extraction, namely the Open Information Extraction (Open IE). An evaluation of several existing Open IE systems shows a good performance on precision. However, improvement is still needed to boost the recall. A relation between entity pair in simple sentence is detected easier by the Open IE system rather than in complex sentence. In this paper, we propose a clause extraction approach employing phrase feature and requiring no learning, focusing on the entity pair. The proposed approach needs less computational cost than the previous work that employing deep parse feature or requiring learning. The experimental result shows that by extracting simpler clause, the performance of Open IE system increases. The average of best F-measure achieved in the evaluation on three benchmark datasets is 0.62, outperforms the previous work.