It's PG's parameters for a naive Bayes model based on these LDA topics, learned by taking his comments on HN as upvotes for the articles content.