基于上下文环境和句法分析的蛋白质关系抽取

摘要

Considering the one-sidedness of features used in many Protein-Protein Interaction (PPI) extraction methods, a new approach was proposed to extract rich features from context information and syntax structure for PPI extraction. Various features, such as lexicon, position, distance, dependency syntax and deep syntax features constitute feature set, and the Support Vector Machine (SVM) classifier was used for PPI extraction. The experimental evaluation on multiple PPI corpora reveals that the rich features can utilize more comprehensive information to reduce the risk of missing some important features. This method achieves state-of-the-art performance with respect to comparable evaluations, with 59. 2% F-seore and 85. 6% Area Under Curve (AUC) on the Aimed corpus.%针对蛋白质交互作用关系( PPI)抽取方法中特征利用的片面性问题,提出了一种从上下文环境和句法结构中抽取特征的方法.该方法抽取词法特征、位置特征、距离特征、依存句法特征和深层句法特征等丰富特征构成特征集,并且使用支持向量机( SVM)分类器进行PPI抽取.方法在5个公开的PPI语料上进行了评估.实验结果表明,丰富特征有效地利用了更为全面的信息,避免丢失重要特征的危险,得到了较好的PPI抽取性能.即在AImed语料上的实验取得了59.2％的F值和85.6％的曲线下面积(AUC)值.