For my site, I coded a recommendation engine based on cosine similarity. I initially tried Bayes' Theorem and conditional probabilities but abandoned the effort when I didn't see results that made a lot of sense (probably due to a lack of training data). Instead, I used https://github.com/aoiaoi/CosineSimilarity as a base and then modified. If the results prove effective over time, I'm going to make the algorithm an option for users to select for generating their front page instead of upvotes.