Methodology
Bayesian Smoothing (review scoring)
Last updated
A statistical technique that pulls low-volume review averages toward a neutral prior, preventing high-variance ratings from low sample sizes from distorting overall scores.
Full definition
Bayesian smoothing applied to review ratings adjusts each product's average rating by combining it with a neutral prior, weighted by the number of reviews. The formula is: smoothed = (n / (n + m)) × raw + (m / (n + m)) × prior, where n is review count, m is the smoothing weight, and prior is the neutral score. With m=15 and prior=7.0 on a 1-10 scale, a tool needs ~30 reviews before user data substantially moves its score. This protects against the '5-star/5-review' flattery problem common in newly-launched tools and the inverse 'one bad early review' penalty. Used by IMDb (top 250 algorithm), Beer Advocate, and the Content Creator Tools methodology.
Examples
- ·A tool with 12 reviews at 4.2/5 (raw) might have a Bayesian-smoothed score of 7.5/10 after pulling toward the prior, versus 8.0/10 raw — accounting for low confidence in 12 reviews.
- ·Comparing two tools: one with 8000 reviews at 4.0/5 vs one with 50 reviews at 4.3/5. Without smoothing, the smaller-sample tool looks better. With smoothing, the larger-sample tool's signal carries more weight.
- ·Used in the Content Creator Tools final_score formula: final_score = 0.8 × expert_score + 0.2 × bayesian_trustpilot.