【摘要】 在统计自然语言处理中会经常遇到一类参数估值问题,就是当观察数据为不完全数据时如何求解参数的最大似然估计,EM算法就是解决这类问题的经典算法。给出了EM算法的基本框架,结合HMM和PCFG模型给出如何应用EM算法求解参数的极大似然估计,讨论了EM算法的优点和不足之处。更多还原

【Abstract】 n statistical natural language processing,one class problem is often encountered that how to estimate the parameter’s maximum-likelihood estimation when observed data set is incomplete.EM algorithm is the classical method to solve this problem.The basic fra-mework of the EM algorithm is described,and then how to apply the EM algorithm is demonstrated to solve the problem of maximum-likelihood parameters estimation combine with the models of HMM and PCFG.Finally,the advantages and disadvantages of EM algo-rithm are discussed.更多还原