Abstract: As one of the important research topics in information extraction, entity relationship extraction is of great significance to the construction of knowledge graph data layer. This paper proposes a two-stage classification technique based on three-way decisions to extract the entity relationship. Firstly, the SVM three-decisions classifier is constructed to implement the first phase entity relation extraction. The softmax multi-class function is used as a probability function of three-way decisions, Then, the KNN classifier is used to classify the three-way decisions middle domain sample into two-stage classification. According to the corpus of ACE2005 as the experimental data, the results of the three-way decisions two-stage classification are compared with the traditional SVM method. The experimental results show that the two-stage entity relation extraction method based on three-way decisions has achieved good classification effect.

Key words: entity relation extraction, three-way decisions, Support Vector Machine（SVM）, K-Nearest Neighbor（KNN）, softmax function

摘要： 实体关系抽取作为信息抽取研究的重要研究课题之一，对知识图谱数据层的构建有着重要的意义。提出一种基于三支决策的两阶段分类技术实现实体关系抽取，首先构建SVM三支决策分类器实现第一阶段实体关系抽取，采用softmax多分类函数作为三支决策概率函数，然后采用KNN分类器对三支决策分类后的中间域样本进行二阶段分类。以ACE2005的语料作为实验数据，将三支决策两阶段分类结果与传统SVM方法分类结果进行比较，实验结果表明，基于三支决策的两阶段实体关系抽取方法取得了很好的分类效果。

关键词: 实体关系抽取, 三支决策, 支持向量机（SVM）, K最近邻（KNN）, softmax函数