Abstract: Research on non-taxonomical relations from domain concepts is an important task in ontology learning. This paper presents an acquiring method for non-taxonomical relations of special domain based on unsupervised learning. This method employs association rules to get the concept pairs of special domain, regards the high frequency verbs of the concept pairs as the candidate non-taxonomical relations labels, non-taxonomical relations labels is determined by VF*ICF metrics, use log likelihood ratio to assign the non-taxonomical relations labels to corresponding domain concepts pairs. Experimental results demonstrate the increment of the accuracy and completeness of the non-taxonomical relations extraction.

Key words: ontology learning, unsupervised learning, association rules, non-taxonomical relations, VF*ICF metrics, Log likelihood ratio

摘要： 领域概念非分类关系的获取是本体学习的一项重要任务，提出了一种基于非监督学习的非分类关系自动获取方法。该方法首先通过关联规则获取特定领域概念对，然后将概念对之间的高频动词作为候选的非分类关系标签，接着利用VF*ICF度量法来确定非分类关系标签，最后通过对数似然比评估方法将得到的非分类关系标签分配给对应的领域概念对。实验结果表明该方法可以有效提高非分类关系抽取的准确率和召回率。

关键词: 本体学习, 非监督学习, 关联规则, 非分类关系, VF*ICF度量法, 对数似然比