We really like the Tinder swiping feature engaging user experience and how it provides better recommendations based on user's past preferences through swiping "like" or "dislike". And members in the team work in Storefront and have dealt with daily sales features.
Therefore, we came up with this idea combining Tinder user experiecne with Wayfair's product sales.
What it does
The web app has a similar interface and experience to tinder continuously displaying photos of various furniture cards one after another in a stack. Basing on the user's preference swiping "like" or "dislike", we inject the data into our deep learning model to provide products accurately fitting the user's taste.
The app also provides a limited 2 minute extra sales off on the current card being displayed such that user can have some prominent extra savings if they check out right now. Thus encouraging user's purchase behavior.
How We built it
We divide our team in front end and back end. Brandon was responsible to crawl public data and built product objects in the backend. The backend also handles the deep learning process and send recommended json objects to the front end.
On the front end, Zhengbang and Neha built the app with AngularJS and its Swing framework.
Challenges We ran into
- Jquery Plugin isn't compatible with angularJS so we have to built directives
- unsupervised deep learning is hard to built with good accuracy
- Not enough user data to train the model
- Button clicking and the swiping behavior is hard to be coherent
Accomplishments that I'm proud of
- Built a deep learning based recommendation engine
- Great mockup for UI
- Great UX on desktop and mobile
What I learned
- Collaborating backend and frontend
- AngularJS directives and services
- Optimizing User Experience based on feedback
What's next for Wayfair Match
- Model specific for each user (we only have one user model now)
- Keep training the model
- Incorporating CMS data so the product recommended is dynamic