摘要： 提出一种实体关系抽取方案，该方案根据实体关系抽取的特点，在常用特征基础上新增语义角色特征用于构建特征向量，并利用支持向量机构造分类器。在SemEval-2010评测任务8提供的数据上进行实验，在判断候选实体对的关系类型上F1值达到81.6%，与未加入语义角色特征相比提高4%，结果表明该方案语义角色特征有利于实体语义关系抽取。

关键词: 关系抽取, 语义角色特征, 支持向量机, 特征向量, 分类器

Abstract: This paper proposes a new entity relation extraction method with a variety of common features plus semantic role features according to the characteristics of entity relation extraction and using the feature vector to build a SVM-classifier. When judging the relation type, this method achieves the F1 measure of 81.6% which is increased by 4% compared with no semantic role features on the data supplied by the evaluation task eight of SemEval-2010. Experimental results show that semantic role feature is benefit to entity relation extraction.

Key words: relation extraction, semantic role features, Support Vector Machine(SVM), feature vector, classifier

中图分类号: 

• TP18