Acquisition of Part-Whole Relations Based on Unsupervised Learning

An unsupervised learning method was proposed to solve the problem of part-whole relation extraction from Chinese free texts. A subsequence extraction algorithm was firstly introduced that can acquire concept pairs and their context patterns from domain texts, and a distributional semantic model was constructed according to concept pairs and context patterns of concept pairs. Then a co-clustering algorithm was applied to group the concept pairs with the same semantic relations together. L1 regularized logistic regression model was trained to select clustering feature and obtain the context pattern which represents semantic relation of each cluster. At last, according to the patterns, the clusters expressing part-whole relation were identified and part-whole relation concept pairs were acquired. The experimental results indicate the proposed method is effective and its F measure is up to 68.97% which is superior to the traditional clustering (55.77%) and pattern matching methods(61.95%).