文摘
This paper proposes a bootstrapping-based method to extract multiple entity relations. Compared with previous entity relation extraction methods, this method analyzes the syntax and semantics of sentences based on traditional context pattern representation. In this way, the features of keyword with the nearest syntactic dependency, phrase structure distance and semantics are extracted so as to form new semantic patterns. To reduce the noise caused by pattern extension, patterns and instances are adopted to verify their reliability mutually. In addition, by combining the information entropy of patterns, accurate and significant instances are selected. Experimental results show that this method effectively improves the quality of patterns and obtains favorable extraction results.