I work with Ranjitha Kumar’s Data Driven Design Group. One of her group’s projects is designing the future of personal fashion via machine learning. Within this project, I’m working on an aspect called “fashion justification.” Anyone can
give fashion advice to a user, but how will the user believe in your recommendations? How can we establish trust? One way is by generating coherent explanations to accompany recommendations. In self driving vehicles, users feel much more
comfortable understanding why exactly the car decided to drive forward ("because there are no other cars in its lane”), for example. The same can be said about fashion recommendations. Users often wonder, ‘why would this particular dress
look good on me?’
Fashion has—and always will be—super personal. Thus, the more tailored the explanation is to the user, the better. Perhaps the user as a certain body shape or hair color, and they want their clothing to flatter those attributes. Or
perhaps they’re attending a wedding in Spain, and want to look the part. While it is easy for a human, personal stylist to understand specific requests and infer physical attributes for the client (through face-to-face interaction,
observing the client’s lifestyle, etc), it is much harder for a computer to do so.
Our goal is to overcome this gap and allow a program to produce personalized fashion justifications to users. We already have a database of thousands of key words we can scrape from clothing sites. They range from observable features
(color, material, silhouette) to notes about style (classy, professional, etc). We aim to map these key words to a host of more personalized, organic explanations about how these clothing features could flatter a user’s appearance, or be
appropriate for their big event. Then, when these keywords are detected, the program will be able to piece together these snippets and create a complete explanation. Ultimately, we hope to implement this via our own search engine.
In addition to this main project, I am also learning a host of skills (building web crawlers, learning about cloud computing, web programming) through weekly tutorials with the other researchers.
View the final report here.