词边界字向量的中文命名实体识别  

Chinese named entity recognition via word boundary based character embedding

作　　者：姚霖[1,2,3] 刘轶[1] 李鑫鑫[4] 刘宏[2] 

机构地区：[1]深港产学研基地,广东深圳518057 [2]北京大学信息科学技术学院,北京100871 [3]哈尔滨工业大学软件学院,黑龙江哈尔滨150001 [4]哈尔滨工业大学深圳研究生院计算机科学与技术学院,广东深圳518055

出　　处：《智能系统学报》2016年第1期37-42,共6页CAAI Transactions on Intelligent Systems

基　　金：原创项目研发与非遗产业化资助项目(YC2015057)

摘　　要：常见的基于机器学习的中文命名实体识别系统往往使用大量人工提取的特征,但特征提取费时费力,是一件十分繁琐的工作。为了减少中文命名实体识别对特征提取的依赖,构建了基于词边界字向量的中文命名实体识别系统。该方法利用神经元网络从大量未标注数据中,自动抽取出蕴含其中的特征信息,生成字特征向量。同时考虑到汉字不是中文语义的最基本单位,单纯的字向量会由于一字多义造成语义的混淆,因此根据同一个字在词中处于不同位置大多含义不同的特点,将单个字在词语中所处的位置信息加入到字特征向量中,形成词边界字向量,将其用于深度神经网络模型训练之中。在Sighan Bakeoff-3(2006)语料中取得了F189.18%的效果,接近当前国际先进水平,说明了该系统不仅摆脱了对特征提取的依赖,也减少了汉字一字多义产生的语义混淆。Most Chinese named entity recognition systems based on machine learning are realized by applying a large amount of manual extracted features. Feature extraction is time-consuming and laborious. In order to remove the dependence on feature extraction,this paper presents a Chinese named entity recognition system via word boundary based character embedding. The method can automatically extract the feature information from a large number of unlabeled data and generate the word feature vector,which will be used in the training of neural network.Since the Chinese characters are not the most basic unit of the Chinese semantics,the simple word vector will be cause the semantics ambiguity problem. According to the same character on different position of the word might have different meanings,this paper proposes a character vector method with word boundary information,constructs a depth neural network system for the Chinese named entity recognition and achieves F189.18% on Sighan Bakeoff-32006 MSRA corpus. The result is closed to the state-of-the-art performance and shows that the system can avoid relying on feature extraction and reduce the character ambiguity.

关 键 词：机器学习 中文命名体识别 深度神经网络 特征向量 特征提取 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]