Latent Dirichlet Allocation

4.3 Probabilistic latent semantic indexing

1. What is the pLSI model, and why isn’t it a well-defined generative model of documents?
   \(p(d, w_n) = p(d) \sum_z p(w_n \mid z)\, p(z \mid d)\): a document index \(d\) and a word are independent given a topic. But \(d\) is a dummy index over training documents, so \(p(z \mid d)\) exists only for them. LDA instead treats topic proportions as a \(k\)-parameter hidden random variable, so parameters don’t grow with the corpus.
   • Blei et al., 2003, Latent Dirichlet Allocation (Journal of Machine Learning Research), §4.3, pp. 1000–1001