I'm looking to quickly get smart on how to add personalization into a Bayesian-based recommendation system.

I'm using clickstream data and Bayesian statistics to estimate probabilities of purchase to rank a list of products on. How do I incorporate user-specific history to modify my generic probabilities for a specific user? For example, a user clicked a product in a previous site visit. How do I estimate the purchase probability during a successive visit? What is the standard approach here?

Looking for papers, blogs, people that can help illuminate.


1 Answer 1


Look into using Collaborative Filtering (CF) via Matrix Factorization in machine learning to get personalization. You can use CF w/ the Bayesian Personalized Ranking (BPR) cost function if you want to keep the Bayesian connection. See this paper: https://arxiv.org/abs/1205.2618 and this intro into recommender system using machine learning: http://www.eggie5.com/99-recommender-systems


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.