Chinese Entity Relation Extraction Based on Syntactic Features

Abstract

Entity Relation Extraction (ERE) is an important research topic in the field of information extraction. However, to the best of our knowledge, only a few ERE works have been done for Chinese corpus. Because the syntactic features of Chinese sentences and English sentences are very different, existing algorithms for English corpus cannot be directly applied to Chinese corpus. Thus, in this paper, we propose a novel Chinese entity extraction system based on syntactic features (named SF-CERE). The basic idea of SF-CERE is given as follows. Firstly, we extract candidate relation triples based on verbs and verb-nouns as relation keywords to avoid pre-defining relation types. Secondly, the triples are filtered using the positional constraints between relation keywords and entity pairs. Thirdly, we summarize four major Chinese syntactic features to expand the identified relation triples and improve accuracy. Finally, we use the method of relation transfer to mine and infer implicit relation triples. The experimental results on two real-world dataset (i.e., the encyclopedia dataset and the news dataset) show that SF-CERE effectively improves the quality of the relation triples and obtains good extraction performance.