Machine Learning Algorithms

Probabilistic latent semantic analysis

The previous model was based on a deterministic approach, but it's also possible to define a probabilistic model over the space determined by documents and words. In this case, we're not making any assumption about Apriori probabilities (this will be done in the next approach), and we're going to determine the parameters that maximize the log-likelihood of our model. In particular, consider the plate notation (if you want to know more about this technique, read https://en.wikipedia.org/wiki/Plate_notation) shown in the following figure:

We assume we have a corpus of m documents and each of them is composed ...