【摘要】 针对标准EM算法在汉语分词的应用中还存在收敛性能不好、分词准确性不高的问题,本文提出了一种基于极大似然估计规则优化EM算法的汉语分词认知模型,首先使用当前词的概率值计算每个可能切分的可能性,对切分可能性进行"归一化"处理,并对每种切分进行词计数,然后针对标准EM算法得到的估计值只能保证收敛到似然函数的一个稳定点,并不能使其保证收敛到全局最大值点或者局部最大值点的问题,采用极大似然估计规则对其进行优化,从而可以使用非线性最优化中的有效方法进行求解达到加速收敛的目的。仿真试验结果表明,本文提出的基于极大似然估计规则优化EM算法的汉语分词认知模型收敛性能更好,且在汉语分词的精确性较高。更多还原

【Abstract】 In view of bad convergence and inaccurate word segmentation of standard EM algorithm in Chinese words segmentation, this paper put forward a cognitive model based on optimized EM algorithm by maximum likelihood estimation rule. Firstly, it uses the probability of current word to calculate the possibility of each possible segmentation and normalize them. Each segmentation is counted by words.Standard EM algorithm cannot make sure converging to a stable point of likelihood function, and converging to a global or local maximum point. Therefore, the maximum likelihood estimation rule is adopted to optimize it so as to use an effective method in nonlinear optimization and accelerate the convergence. the simulation experiments show that the optimized EM algorithm by maximum likelihood estimation rule has better convergence performance in the Chinese words cognitive model and more accurate in the words segmentation.更多还原