一种基于树核函数的半监督关系抽取方法研究  

A semi-supervised method based on tree kernel for relationship extraction

作　　者：刘晓勇[1] 

机构地区：[1]广东技术师范学院计算机科学学院,广东广州510665

出　　处：《山东大学学报（工学版）》2015年第2期22-26,32,共6页Journal of Shandong University（Engineering Science）

基　　金：广东高校优秀青年教师培养计划资助项目(Yq2013108)

摘　　要：为了解决传统的半监督关系抽取算法易产生的"语义变异"问题,提出一种新的基于树核函数的半监督关系抽取算法。该算法主要采用树核函数和种子集约束扩展两个策略,弱化"语义变异"现象带来的关系抽取不够准确的问题,提高关系识别的正确率。在基准数据集Pop Bank上的试验研究表明,提出的使用约束机制扩充种子集的半监督学习方法在4个评价指标上(Precision,Recall,F-measure,Accuracy)均优于常用的两种关系抽取方法,从而验证了该算法与其他算法相比能够具有较好的关系抽取能力。It was difficult for traditional semi-supervised relation extraction methods to solve ＂semantic variation＂ prob- lem. A new semi-supervised relation extraction algorithm based on ensemble learning was prorosed and named L-EC- RE, which used two strategies, one was tree kernel and the other was constrained extension seed set. Experimental study on PopBank benchmark data sets showed that L-EC-RE had better performance than two usual relation extraction algorithms in four assessment criteria, which were Precision, Recall, F-measure and Accuracy.

关 键 词：关系抽取 树核函数 支持向量机 半监督方法 语义变异 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]