1.Dirichlet-Multinomial共轭
2.LDA=PLSA+bayes
3.训练方法有gibbs sampling(吉布斯采样)和variation inference-EM算法(变分推断-EM算法)
4.LDA将document-term矩阵分解成topic-term矩阵和document-topic矩阵
latent dirichlet allocation
原文:https://www.cnblogs.com/pandas-blue/p/11745612.html