navgupta14 / DiscoverRestaurants

Personal restaurants recommendations using Discover Privilege offers

DiscoverRestaurants FAQ

What is the problem we're solving?

Users want to find Diner's club merchants, but have no metadata to inform their decision, unless they go and get it themselves.

What is our solution?

Bring the best available metadata on Diner's club merchants to the user in the web app.

How did we do this?

Using a ML model, we create a curated list of merchants for each user and sort that list
according to each individual user's predicted preferences.
We use the following APIs:
- Discover City Guides
- Yelp

What challenges did we face?

- Handling pagination in TamperMonkey script
- Yelp API timeout preventing model application to ALL merchants
- Cold start for new users: without existing spend history, the model will have to compensate
  by using proxy factors such as age, home address,  income, credit score, etc.

How does the ML model work?

- Our model uses Yelp rating, # of reviews, whether a DCI privilege exists, and categorical 
 spend history to produce a score that indicates whether the user would prefer any given merchant.
 - Higher score means higher predicted user preference
- The model then sorts the merchants by score (descending) to deliver a curated list of merchants.

Why is this important?

- By providing a curated list of merchants, DCI will improve User-experience and drive usage of the app.

What did Technologies did we use?

### Backend
- AWS Lamda
- AWS DynamoDB
- AWS API Gateway

### Machine-Learning and Data Libraries
- NumPy
- SciPy
- Pickle

### Frontend
- TamperMonkey