基于最大熵方法的统计语言模型  

A Statistics Language Model Based on the Maximum Entropy Ap proach

作　　者：徐延勇[1] 郭忠伟[1] 周献中[1] 

机构地区：[1]南京理工大学自动控制系,南京210094

出　　处：《计算机工程与应用》2002年第5期53-55,121,共4页Computer Engineering and Applications

基　　金：国防科工委跨行业基金项目资助

摘　　要：针对现有统计语言模型中存在计算量过大和系统负担过重的问题,该文提出了一种基于最大熵方法的统计语言模型。模型在参数估计阶段,引入约束最优化理论中拉格朗日乘数定理和牛顿迭代算法,以确保模型在多个约束条件中可求出最优化参数值;在特征选择阶段,采用计算近似增益的平行算法,解决模型计算量过大和系统开销问题。将该模型用于汉语句子分析的软件实验中表明:模型具有较高的计算效率和鲁棒性。To solve the problem of computational expensiveness and system spending of the statistics language model in existence,a sort of mathematic algorithms of the statistics language model based on the maximum entropy approach is described detailedly in this article,In parameter estimation stage of the model,Lagrange multipliers from constrained optimization theory and the Newton iterative scaling algorithm are applied to get the optimal parameter values in the more constraint condition;In the feature selection stage,the computing approximate gains in parallel is adopted in order to solve the computational expensiveness of the model and system spending.Software experiments about the Chinese sentence parsing show the model presented in this paper has higher efficiency and robustness.

关 键 词：自然语言处理 统计语言模型 最大熵方法 鲁棒性 计算机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]