基于向量相似度计算的半监督的名实体识别  

Named entity recognition of semi-superivised based on vector similarity

作　　者：谭红叶[1] 赵铁军[1] 王浩畅[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院

出　　处：《计算机工程与设计》2008年第19期5047-5050,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60575041、60473139、60775041);国家863高技术研究发展计划基金项目(2006AA01Z150);山西省青年科技基金项目(20051018)

摘　　要：提出一种基于向量相似度计算的半监督的NER方法,主要思想是:首先利用bootstrapping方法获取NER所需的各种特征;然后将待测实例表示为实例特征向量,每一类名实体表示为类特征向量;最后根据每个类特征向量与实例特征向量的相似度进行分类。在人民日报语料上选取疾病名、武器名、交通工具名进行相关测试,F测度分别为:77.4%,66.1%和73.1%,结果令人满意。A semi-supervised method for NER based on vector similarity is presented. The main ideas include： firstly, the features of NE are obtained based on the bootstrapping algorithm, then every instance to be recognized is represented as a feature vector, and so does each category of NE, finally, the vector similarity between the two kinds of vectors is computed, and according to the vector similarity, the final category is decided. The experimental results of disease, weapon and vehicle names on the people daily corpus show that the performances are satisfying, with the F-measure of 77.4%, 66.1% and 73.1% respectively.

关 键 词：名实体识别 特征向量 向量相似度 半监督学习 自举 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术]