融合迁移学习的中文命名实体识别  

Research on Chinese Named Entity Recognition Fusing Transfer Learning

作　　者：王红斌[1] 沈强[1] 线岩团[1] 

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650504

出　　处：《小型微型计算机系统》2017年第2期346-351,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61462054;61363044)资助;云南省科技厅面上项目(2015FB135)资助;云南省教育厅科学研究基金重点项目(2014Z021)资助;昆明理工大学省级人培项目(KKSY201403028)资助

摘　　要：命名实体识别是自然语言处理研究领域中的一项很重要的基础性任务,是实体关系抽取和事件抽取等高层任务重要基石.如何在缺乏标注语料或只有少量标注语料条件下,提高命名实体识别的性能是自然语言处理领域的一个重要研究方向.针对这一问题,提出一种基于实例的迁移学习算法——TLNER_AdaBoost.该方法通过自动调整训练集中实例样本的权重和计算辅助训练样本的迁移能力来提高训练语料质量,并选取采用不完全标注语料的自学习方法和采用完全标注语料的基于条件随机场的方法来对该方法进行实验对比分析.经实验对比分析得知,本文方法在提高命名实体识别的准确率、召回率和F值的同时,大大降低了人工标注语料的工作量.Named entity recognition is a very important basic task in natural language processing research;it is the high-level task cornerstone of the entity relationship extraction and event extraction. When we are faced with the lack of annotation corpus, or only a small amount of tagging corpus, how to improve the performance of named entity recognition was an important research direction in natural language processing. In order to solve this problem, we present an instance-based transfer learning algorithm TLNER_Ada- Boost. This method improved the quality of training corpus by adjusting automatically weight of the whole training instances samples and calculated the transfer ability of assist training samples, then improved the performance of named entity recognition. We use the self-learning method with incomplete tagging corpus and conditional random field method with completely tagging corpus as the comparison method to do experiment contrast analysis. The experiment results show that our method can not only improve the rate of accuracy, recall and F-Measure of named entity recognition, but also decrease the workload of manual tagged corpus.

关 键 词：命名实体识别 迁移学习 机器学习 TLNER_Ada BOOST 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]