摘要： 基于标签传播的半监督学习算法能够提升少量标注数据下的关系抽取效果,但是随机选择训练样本会使关系抽取性能降低。为了从海量的网络信息中提取出可靠性较高的人物关系,将标签传播算法与主动学习相结合用于人物关系抽取。在训练数据获取中,主动选择不确定性最大的样本进行标注。在人物关系上的实验结果显示,主动学习方法的引入可使平均F1值比标签传播算法提升2.3%。

关键词: 人物社会关系, 特征提取, 标签传播, 主动学习, 关系抽取, 半监督学习

Abstract: In order to extract personal relations of high reliability from the mass network information,the semi-supervised learning algorithm based on label propagation can improve the performance of relation extraction under small amount of labeled data,but randomly selecting training sample may cause the reduction of the relation extraction performance.This paper combines label propagation algorithm and active learning so as to extract the relationship between the characters.In the training data acquisition,the maximum uncertainty of the sample is actively selected for label.Experimental results on personal relation show that the active learning method improves the average F1 by 2.3% than label propagation algorithm.

Key words: personal social relation, feature extraction, label propagation, active learning, relation extraction, semi-supervised learning

中图分类号: 

• TP393