Inspiration
The biggest (and perhaps only) downside of Wayfair's immense inventory comes down to navigation. The notion of sifting through 7 million-plus products and an ever-expanding list of services can be a little overwhelming to the average user and especially so to our older-skewing target demographic. Simply put, Wayfair needs a Siri (or something like it) to help our customers find what they need, answer any questions they may have, and perhaps one day, even offer them style or product tips! Our engineer, Jack, was struck with this realization after having his own little tussle with the site, and WayFriend was born!
What It Does
WayFriend is a chatbot that helps you navigate the Wayfair website, find specific products, get started on a return, or point you to one of our many services. It can even tell you a joke! Using a third-party, open-source algorithm to analyze the general sentiment of a user's language, WayFriend can determine whether a customer is speaking positively, neutrally, or negatively and respond accordingly with both language and emojis.
How We Built It
WayFriend uses three different data science algorithms:
- Sentiment Analysis Algorithm ( https://github.com/JWHennessey/phpInsight )
- Product Classification Algorithm (proprietary)
- Profanity Detection ( https://github.com/raymondjavaxx/swearjar-php )
On top of that, we built our own algorithms designed to point users to specific aspects of the site (ie. Room Ideas, Registry, etc.) by building a bank of "custom analyzer terms" that triggered the bot to route the user to an appropriate landing page, as well as a bank of "situational responses" that allows the bot respond to a given user sentiment.
Challenges We Faced
The data science algorithm was not accurate at first, so we had to apply our own algorithm onto it in order to make it feasible. We also had to replace and ignore some words that were causing our algorithm to misidentify user messages (ie. replacing "couch" with "sofa" to better match our catalog data). Our bot also ran so smoothly that we were forced to slow it down in order to improve the user experience and simulate the chatbot thinking/typing.
Accomplishments We're Proud Of
We are most proud of WayFriend's ability to understand user's emotions and gracefully direct them to the section of the website, as well as its sleek, seamless interface. WayFriend also handles profanity laced messages from angry (as well as excited) customers in a polite way, hopefully leading to increased customer retention. Wayfriend's friendly personality should be a huge hit among customers!
What We Learned
We learned how to take free text from a user and classify it in various different ways regarding sentiment, profanity, relational data. We also learned a lot about natural language processing and the challenges associated with integrating that into a large, living system.
What's Next for WayFriend
- Consult with Data Science to improve all our algorithms for accuracy.
- Work with copywriters to enhance the content/personality of the bot.
- Add feedback for the bot and work with user testers to improve the overall functionality.
- Integrate the bot with recommendations and metadata about products in order to answer any question the user could conceivably have (ie. make it able to answer product-specific questions and give tips to users based on previous purchases)
- Integrate the bot into our Twitter/social posts.
The Team
We are three friends who met in college and teamed up for this Hackathon!
Source Code
https://git.csnzoo.com/wayfair/resources/commits/wayfriend https://git.csnzoo.com/wayfair/php/commits/wayfriend
(only viewable by internal Wayfair folks)
Log in or sign up for Devpost to join the conversation.