Hands-On Machine Learning for Algorithmic Trading

Reverse-engineering the process

The generative process is fictional but turns out to be useful because it permits the recovery of the various distributions. The LDA algorithm reverse-engineers the work of the imaginary author and arrives at a summary of the document-topic-word relationships that concisely describes the following:

• The percentage contribution of each topic to a document
• The probabilistic association of each word with a topic

LDA solves the Bayesian inference problem of recovering the distributions from the body of documents and the words they contain by reverse-engineering the assumed content generation process. The original paper uses variational Bayes (VB) to approximate the posterior distribution. Alternatives include ...