Extracting Part-Whole Relations from Unstructured Chinese Corpus

An important problem in text mining is the automatic extraction of semantic relations. The paper provides a domain independent method for automatic extraction of part-whole relations in Chinese corpusa. The method consists of there phases. First, a set of lexico-syntactical patterns for part-whole relations are designed using known pairs of concepts encoding part-whole relations as seeds, and manually filtering the extracted sentences. Second, Pairs of concepts are extracted using the patterns from a training corpus, which may reflect part-whole relations. Finally, the extracted pairs of concepts are further confirmed using a set of heuristic rules generated based on an analysis of Chinese syntactical and semantic features. Based on a test corpus, the method achieves satisfactory results.