[EMNLP2015]Distant supervision for Relation Extraction via Piecewise Convolutional Neural Networks

paper中提高两个概念:
(1)distant supervision: 如果两个实体在已知知识库中有关系,则包含这两个实体的中,两个实体的的关系默认就是知识库中给定的关系, 因为人工标注语料代价大,所以一般用distant supervision标注语料
(2)multi-instance learning:训练语料是很多带有正负标签的包,每个包里有很多示例,如果一个包里包含至少一个示例是正标签则这个包就标记为正,如果一个包里的示例都是负的,则该包的标签就是负的,包里具体的示例是什么标签并不明确,能确定的就是包的标签,在测试语料中,我们的目的是给包打标签
[EMNLP2015]Distant supervision for Relation Extraction via Piecewise Convolutional Neural Networks
[EMNLP2015]Distant supervision for Relation Extraction via Piecewise Convolutional Neural Networks
[EMNLP2015]Distant supervision for Relation Extraction via Piecewise Convolutional Neural Networks