Bootstrapping-based relation extraction in financial domain
Bing Kong, Ruifeng Xu, Dongyin Wu · 2015
Relation extraction plays an important role in many natural language processing tasks, such as knowledge graph and question answering system. This paper presents a novel method to extract relation from Chinese financial news by incorporating relation pattern matching and bootstrapping based pattern expansion. The seed patterns are firstly manually compiled. They are applied to matching the sentences from unlabeled text The new patterns are then discovered through finding the maximum common substring sequences between the sentences to generate candidate patterns and estimating the quality of candidate patterns. The pattern lib is expanded iteratively. These patterns are applied to running Chinese text in finance domain for extracting the target relations. Experimental results show that our proposed relation extraction method achieves good performance with few labeled data.