一种融合语义角色和依存句法的实体关系抽取算法

An entity relation extraction algorithm based on semantic roles labeling and dependency parsing

针对目前基于语义角色的实体及关系抽取技术效果不理想以及存在无法正确提取多个宾语的缺陷,提出一种基于语义角色和依存关系融合的方法:1)通过语义角色标注的方式抽取主语和核心谓词;2)以核心谓词为切入点,通过依存句法关系分析句子的并列结构(COO)和动宾结构(VOB),抽取其中的宾语实体;3)整合主语、核心谓词和宾语构成[实体关系实体]三元组。对该算法和单纯依赖语义角色识别的算法进行了对比实验,结果显示该算法的精确率、召回率、F1值3个指标更优,表明这种实体关系抽取的方法可行有效,且在抽取多宾语的任务中效果明显。

The entity and relationship extraction technology based on semantic roles labeling is not effecctive enough to correctly extract multiple objects. This paper proposes a method based on semantic roles labeling and dependency parsing: 1) Extracting subjects and core predicates through semantic roles labeling;2) Taking the core predicate as the entry point, analyzing the COO structure and VOB structure of the sentence through the dependence sentence relationship, and extracting the object entity among them;3) Integrating subject, core predicate and object to form [ Entity Relation Entity] triple entities. The algorithms based on text and semantic role labeling are compared, and the results show that the text algorithm has better accuracy, recall rate and F1 value, which shows that this method is feasible and effective, and the effect is obvious in the task of extracting multiple objects.