摘要
针对现存的基于EM(Expectation maximization)迭代的无指导词义消歧方法收敛缓慢、计算量大的问题,利用互信息和Z-测试结合的方法选取特征,并通过一种统计学习算法估算初始参数值.实验结果表明改进方法有效地提高了汉语词义消歧的准确率,具有良好的扩展性和实用性.
The existing word sense disambiguation methods based on expectation maximization (EM) unsupervised learning need a large amount of computation and converge slowly. To address the problems, an improved method is proposed, which makes use of mutual information theory based on Z-test to select features and uses a statistical learning algorithm to estimate initial parameter values. The experimental result shows that the proposed method improves effectively the precision of word sense disambiguation and has good expansibility and practicability.
出处
《自动化学报》
EI
CSCD
北大核心
2010年第1期184-187,共4页
Acta Automatica Sinica
基金
国家自然科学基金(60773100)资助~~
关键词
词义消歧
无指导学习
特征提取
参数估计
Word sense disambiguation, unsupervised learning, feature extraction, parameter estimation