Compare Product Reviews Quickly

When choosing a restaurant to go for dinner or a product to buy have you ever asked yourself: what is better, a product with 4 stars and 3 reviews or a product with 3 stars and 50 reviews?  Our web app Compare My Ratings will help you making better and faster decisions by translating product reviews from two products into numbers that are easy to understand.


It is all about taking the right decision

Decision making is a balance between two forces: expected gain and risk.  The expected gain tell us how much we are expected to gain on average but it does not tell us how variable the gain will be in a single trial or attempt. Risk is a measure of variability in the actual gain.

In our product selection example above it is clear that the ratings are somehow related to the expected gain and that the number of users that have voted for each product is somehow related to the risk. How to make this relation more precise?

When choosing a product we are effectively asking the question: which product is more likely to be the best? 

We have to consider 3 scenarios:

  1. Both products are equally good
  2. The first product is the best
  3. The second product is the best

Wouldn't it be easy if we had a tool to translate rating scores and counts to the chances that each one of these scenarios is true? That is why we created the app Compare My Ratings!



How does it work?

Translating ratings and counts to probabilities is not that hard if you think about it. After all, probabilities are just disguised counts.

But to see that we have some work to do, consider the following: How many different ways can 4 voters score a book resulting in an average score of 4 stars ? Two easy combinations would be: (1) all 4 voters choose 4 stars and (2) two voters choose 3 stars and 2 voters choose 5 stars.

As we can imagine, when the number of voters is large there are many many combinations. But don't worry, our app will do it cleverly for you!

For the geeks out there: Under the hoods, our app is doing Bayesian Inference using a Multinomial-Dirichlet model.