Abstract: It has a significant guiding significance for developing intelligent?auxiliary diagnosis and treatment of depression to develop automatically ontology learning technology of depression drug ontology based on mass biomedicine literatures. Concepts extraction has an effect on learning relations and axioms, and then decides the quality of ontology learning. However, existing algorithms are only suitable for general fields, not for special and fine grit fields. This paper proposes a method of concepts extraction from depression drug field, which employs log-likelihood ratio and domain correlation function based on traditional domain relevance and domain consensus. The experimental results show that the method can reduce the impact on concepts extraction caused by other related fields of depression, and improve the calculation of domain membership degree for low-frequency terms. Compared with others, the precise and recall has been greatly improved.

Key words: ontology learning, concepts extraction, depression, log-likelihood ratio, domain correlation function

摘要： 开发基于生物医学文献的抑郁症药物本体自动学习技术，对于抑郁症辅助诊疗有着重要的指导意义。概念抽取是面向文本的本体学习的基础。然而，现有的本体概念抽取算法在解决特定、细粒度领域的概念抽取问题时性能较差。借鉴传统的领域相关性及领域一致性的思想，综合使用对数似然比和领域关联函数进行抑郁症药物领域的概念抽取。实验结果表明，该算法能够降低抑郁症其他相关领域对概念抽取的影响，同时改善低频术语的领域隶属度计算，提高了准召率。

关键词: 本体学习, 概念抽取, 抑郁症, 对数似然比, 领域关联函数