Outline

Approach inspired by Bayesian probability is to have an initial belief about the 'true' rating of an item, and use user ratings to update this belief, meaning we will have to quantify an initial rating for each item (using a 5 point scale, as most review systems do).

As such, we require 2 parameters:

• The 'true' default rating of an item, if there are no existing ratings 'R' -> the initial belief

• The weight that we want to give this initial belief, versus subsequent ratings 'W' -> the weight of our initial belief, where belief 'R' is worth W user ratings

Examples

Assume you have W ratings of value R along with any user ratings, and find the average. Given R=2 and W=3, where N is number of user ratings, V is their average

Weighted_Rating = (RW + NV) / (W+N)

1. For 100 user ratings of 4

• Weighted_Rating = (2 ⋅ 3 + 100 ⋅ 4) / (100 + 3) = 3.94☆
• In comparison, a straight average would give us a 4☆ rating.

2. For 3 ratings of 5 and 1 rating of 4

• Weighted_Rating = (2 ⋅ 3 + 1 ⋅ 4 + 3 ⋅ 5) / (3 + 4) = 3.57☆
• Average = 4.75☆

3. For 10 ratings of 4

• Weighted_Rating = (2 ⋅ 3 + 10 ⋅ 4) / (3 + 10) = 3.54☆
• Average = 4☆

4. For 1 ratings of 5

• Weighted_Rating = (2 ⋅ 3 + 5 ⋅ 1) / (3 + 1) = 2.75☆
• Average = 5☆

Usage

You'll need a Google Maps API, which you can get through the Google Cloud Console. After acquiring a key, export it as an environment variable: export GOOGLE_MAPS_API_KEY='your_actual_api_key_here'. You will be able to enter restaurant names into the menu.

Installation:

git clone https://github.com/your-username/bayesian-food-ratings.git
cd bayesian-food-ratings
python app.py

Todo

• Simplify interface (just a plain textbox in the middle of the page)
• Add auto-completion like in the Google example
• Add history (keep previous searches visible)
• Add brief explanation & example use case (ex. high rating with few reviews)