Self-Improving Bayesian Content Ranking with Click-Through Feedback

This paper presents a content ranking system developed for GameSpot that uses Bayesian scoring with a click-through rate (CTR) feedback loop to continuously self-improve. The system ranks diverse content types (news, reviews, videos, image galleries) using a two-stage approach: (1) a Bayesian scoring stage with dynamic decay derived from pageview slopes, using prior iteration scores as informative priors, and (2) a feedback loop incorporating CTR (modeled over a betadistribution for regularization) that drives Normalized Discounted Cumulative Gain (NDCG) toward 1.0 against a click-ordered ideal. The system continuously self-corrects toward ideal rankings. This paper describes the design, the mathematical framework, and discusses how this approach relates to techniques now standard in recommendation systems.