实体词语义信息对中文实体关系抽取的作用研究

发布时间：2018-11-10 10:14

【摘要】：为了探索从外部词典提取的实体词语义信息在中文实体关系抽取上的作用,提出了知网语义树和《同义词词林》编码树两种新特征,并连同已有的最短路径依存树、知网第一基本义原、《同义词词林》编码、实体大类和实体小类特征一起探究了其对抽取性能的影响。实验数据表明,作为单一特征时,提出的两种新特征性能虽然不如实体大类和实体小类特征,但是比最短路径依存树、知网第一基本义原、《同义词词林》编码好;作为组合特征时,最短路径依存树和两种新特征的组合特征取得了最好的性能。由此可以得出结论,从非语料中获取的最短路径依存树、知网语义树和《同义词词林》编码树特征可以代替从语料中获取的实体大类和实体小类特征,可以用于下一步开放域的关系抽取。
[Abstract]:In order to explore the role of the semantic information of entity words extracted from external dictionaries in the extraction of Chinese entity relations, two new features, the semantic tree of knowledge web and the coding tree of synonym forest, are proposed, together with the existing shortest path dependency tree. The first basic semantic source, synonym forest coding, entity large class and entity subclass feature are studied together to study its influence on extraction performance. Experimental data show that, as a single feature, the performance of the proposed two new features is not as good as that of the entity class and the entity subclass, but it is better than the shortest path dependency tree, the first basic semantic element of the network, and the encoding of the synonym forest. As a combination feature, the shortest path dependency tree and the combined features of two new features achieve the best performance. It can be concluded that the features of the shortest path dependency tree, the semantic tree of the known web and the coding tree of the synonym forest obtained from the non-corpus can replace the features of the large and small classes of entities obtained from the corpus. It can be used for the next open domain relational extraction.
【作者单位】：太原理工大学计算机科学与技术学院;武汉大学软件工程国家重点实验室;
【基金】：武汉大学软件工程国家重点实验室开放课题(SKLSE2012-09-30) 山西省自然科学基金资助项目(2013011015-2)
【分类号】：TP391.1

【相似文献】