摘要
组织机构名等命名实体的识别是信息抽取、机器翻译等任务的重要基础.为了克服识别器训练过程中对标注数据的依赖,本文提出了一种基于主动学习的训练策略,改进了基本的最大熵模型的解码算法和训练过程.实验表明采用主动学习策略的最大熵模型训练算法能够有效减少标注数据的使用.
The recognition of organization names is one of the fundamental tasks in Information Extraction and Machine Translation. To minimize the dependency on the labeled data used to training the model, an improved training strategy based on actire learning is proposed, and the classical recognition system can be elevated rapidly maximum entropy model is extended. The even with small account of selected labeled experiments show that performance of the samples, which prove the efficiency of the active learning training strategies.
出处
《小型微型计算机系统》
CSCD
北大核心
2006年第4期710-714,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60272088)资助
国家"八六三"基金项目(2002AA11401)资助
关键词
主动学习
命名实体识别
最大熵模型
组织机构名
active learning
named entity recognition
maximum entropy model
organization names
