Predicting hypernym–hyponym relations for Chinese taxonomy learning

Hypernym–hyponym (“is-a”) relations are key components in taxonomies, object hierarchies and knowledge graphs. Robustly harvesting of such relations requires the analysis of the linguistic characteristics of is-a word pairs in the target language. While there is abundant research on is-a relation extraction in English, it still remains a challenge to accurately identify such relations from Chinese knowledge sources due to the flexibility of language expression and the significant differences between the two language families. In this paper, we introduce a weakly supervised framework to extract Chinese is-a relations from user-generated categories. It employs piecewise linear projection models trained on an existing Chinese taxonomy built from Wikipedia and an iterative learning algorithm to update model parameters incrementally. A pattern-based relation selection method is proposed to prevent “semantic drift” in the learning process using bi-criteria optimization. Experimental results on the publicly available test set illustrate that the proposed approach outperforms state-of-the-art methods.