This program takes a business name, and uses publicly available data from the Google Places API and the Wikipedia API to construct a “P score” for any given small business, which represents how likely the business is to default on a 3‐year loan we extend to the business.
APIs used:
- Google Places
- Wikipedia
I used these APIs because I wanted to minimize any manual intervention on the part of the user running the script, and many APIs require OAUTH.
I really wanted to use the Facebook Graph API and looked at such modules as python-sdk and facebook-sdk. See [To Do](##To Do) section for more details.
The P Score is an average of two scores, the Google Places score (the score calculated by metrics provided by the Google Places API) and the Wikipedia Score.
The Google Places score is between 0 and 1 depending on these factors:
- rating - the average rating that customers give the business (if not available then 0)
- business_categories - the number of categories this business falls under (i.e. restaurant, airport, etc -- see https://developers.google.com/places/web-service/supported_types). My assumption here was that if the business has many products and industries, then it is less likely to default given that if one of its products is not successful, it can pivot.
The Wikipedia score is between 0 and 1 depending on these factors:
- total_hits - the number of Wikipedia page results when you search the business name
- updated_recently - whether the first page result has been updated within the past year (I used this metric to measure whether the business was still in business)
Get your Google Places API key by clicking "Get a Key" here: https://developers.google.com/places/web-service/
python business_scorer.py "[Google API key] | [Business Name] | [Business Address]"
# These are valid businesses
> python business_scorer.py "AIzaSyDa2cfifSiwMIDSBRWRENhVej5fbuzfweg | Aveda Spa | 598 Broadway, New York, NY"
P score: 0.5
> python business_scorer.py "AIzaSyDa2cfifSiwMIDSBRWRENhVej5fbuzfweg | Home Depot | 40 W 23rd St, New York, NY 10010"
P score: 0.845
> python business_scorer.py "AIzaSyDa2cfifSiwMIDSBRWRENhVej5fbuzfweg | Home Depot | Louisville"
P score: 0.845
# This is an invalid business
> python business_scorer.py "AIzaSyDa2cfifSiwMIDSBRWRENhVej5fbuzfweg | Miranda's Hardware Store | Louisville"
P score: 0.0
As you can see from the above, business name and address do not need to be spelled perfectly because this code uses search-oriented APIs
- I wanted to use the Facebook Graph API Page endpoint so that, given a business's Facebook page ID (the id in the url of its Facebook page, i.e. cocacola), I could access the business's following information:
- is_permanently_closed
- can_checkin && checkins - Number of checkins at the business
- fan_count
- founded
- rating_count
- app_links (app ids for mobile apps)
Given the downsides of oauth, I decided not to use the Facebook Graph API, but may add this functionality in the future.
-
The score is very forgiving and bases its score on returned results, so I would want to make it less forgiving/have fewer false positives.
-
Tests.