摘要： 从生物医学文献中自动地抽取蛋白质相互作用(PPI)关系是文本挖掘的一项重要任务。考虑到特征和分类器的选择对于PPI任务的重要性,提出一种基于丰富特征和多分类器融合的蛋白质关系抽取方法。选取15种词法、句法及语义特征,融合3种分类器,采用文档级别 的10倍交叉验证方法,在5个公开的PPI基准语料上进行评估实验,结果表明,该方法在AIMed语料上取得的F值和AUC值分别为63.7%和87.8%,具有良好的抽取性能。

关键词: 蛋白质相互作用关系抽取, 丰富特征, 支持向量机, 最大熵, 图核

Abstract: Automatically extracting Protein-protein Interaction(PPI) from biomedical literature is a significant task in text mining.Considering the choice of features and the selection of classifier is of great importance for Protein-protein Interaction Extraction(PPIE) task,this paper proposes a method to combine various features and multiple classifiers.Fifteen lexical,syntactic and semantic features,three kinds of classifiers and the standard ten-fold document level cross-validation evaluation method are used to evaluate on the five public PPI corpuses,and results show that the method achieves the preferable F-score(63.7%) and AUC-score(87.8%) on the AIMed corpus which is on the top of the PPI extraction task,and it has better extraction performance.

Key words: Protein-protein Interaction Extraction(PPIE), rich features, Support Vector Machine(SVM), maximum entropy, graph kernel

中图分类号: 

• TP391