Latent Dirichlet Allocation (LDA) is a generative probabilistic model used in natural language processing for topic modeling. It posits that each document in a corpus is a mixture of a certain number of topics, and each topic is a distribution over words. The model’s generative process: for each document, draw a distribution over topics (Dirichlet distributed); for each word in the document, randomly choose a topic according to this distribution, then randomly choose a word from that topic’s word distribution. The parameters (topic-word distributions and document-topic mixtures) are latent and are inferred from the data (usually via approximate methods like collapsed Gibbs sampling or variational Bayes). The output is topics (lists of words with probabilities) that hopefully align with human-understandable themes, and topic proportions per document. LDA was a significant advance in text mining, enabling unsupervised discovery of hidden semantic structure in large text corpora.