上下位关系的自动验证是知识获取中的一个关键问题。提出一种基于混合特征的迭代上下位关系验证方法,从语义、语境、空间结构角度,给出一组上下位关系特征,根据抽样数据分析,将所有特征转化为用于验证的产生式规则,利用这些规则对基于模式获取的上下位关系进行循环迭代验证。实验结果说明了该方法的有效性。
Automatic verification of hyponymy relations is a basic problem in knowledge acquisition from text. This paper presents an iterative method of hyponymy verification based on hybrid features. It initially acquires a set of hyponymy features based on semantic, context and space structure of hyponymy. These features are changed into a set of production rules. The candidate hyponymys using Chinese lexico-syntactic patterns from Chinese corpus is verified with multi-features production rules. Experimental results demonstrate good performance of the method.