A term-topic matrix stores the probabilities or weights that indicate how appropriate each vocabulary term is for each latent topic. It lets a topic-mixture model choose words when generating a bag-of-words document. section 4.5.1
In the LDiA process, a document first receives a length and a topic mixture; the matrix then supplies word probabilities for those topics. LDiA estimates the matrix from a collection of documents. section 4.5.1
From corpus to word choices
Imagine a bot discussing several topics. It samples topics for a document, then repeatedly looks up suitable vocabulary terms in the term-topic matrix until reaching the chosen document length. The matrix is computed from tokenized and vectorized documents. section 4.5.1
