An Integrated Approach for Relation Extraction from Wikipedia Texts

Linguistic-based methods and web mining-based methods are two types of leading methods for semantic relation extraction task. By integrating linguistic analysis with frequent Web information, this paper presents an unsupervised relation extraction approach, for discovering and enhancing relations in which a specified concept participates. We focus on concepts described in Wikipedia articles. By making use of the characteristics of Wikipedia and Web corpus, we define a novel distance function and develop a linear clustering algorithm on the combination of two kinds of patterns: dependency patterns from dependency analysis of texts in Wikipedia, surface patterns generated from high redundant information from the Web corpus. The experiments on two different domains demonstrate the superiority of our approach comparing with previous method. In essence, our approach shows how deep linguistic features contribute complementally with Web surface features to generate a broad variety of relations.