Star Rating Systems
Five-star rating systems are the most common method for collecting user feedback on products, services, and experiences. From Amazon to Yelp to Google Maps, these systems distill complex opinions into a simple 1-5 scale. The challenge lies in computing a fair aggregate rating that accounts for both the average and the number of ratings.
A simple average can be misleading: a product with one 5-star review is not necessarily better than one with 100 reviews averaging 4.5 stars. This is where weighted and Bayesian averages become valuable for fair ranking.
Calculation Methods
Bayesian Average
Where C is a confidence parameter (prior weight), m is the prior mean (e.g., 3.0), n is the number of ratings, and R is the simple average. This pulls ratings with few reviews toward the global mean, preventing items with 1-2 five-star reviews from topping the list.
| Scenario | Simple Avg | Bayesian (C=10,m=3) |
|---|---|---|
| 1 review at 5 stars | 5.00 | 3.18 |
| 10 reviews at 5 stars | 5.00 | 4.00 |
| 100 reviews at 4.5 | 4.50 | 4.36 |
Platform Comparison
- Amazon: Uses machine learning to weight reviews by recency and helpfulness
- Yelp: Filters suspected fake reviews algorithmically
- IMDB: Uses Bayesian average with secret parameters for Top 250
Frequently Asked Questions
What is a good average rating?
On most platforms, the average product rating is around 4.0-4.3 due to self-selection bias (unhappy customers are more likely to leave reviews). A rating below 3.5 is generally considered poor.
How many reviews are needed for reliability?
Research suggests 30+ reviews provide a reasonably stable average. Below 10 reviews, the rating is highly volatile and the Bayesian average is much more appropriate.
Why use Bayesian average?
Bayesian averaging prevents gaming by shrinking extreme ratings from few reviews toward the population mean. It is fairer for ranking items with vastly different review counts.