TweakCorps: Retargeting Existing Webpages for Diverse Devices and Users
Led by Former Associate Prof. Scott Klemmer, Computer Science
What if there were a machine-learning algorithm that could identify and label design elements of web pages?
This project combined design and computer science principles to develop a machine-learning algorithm to label web pages. The algorithm takes the structure of pages, conducts design mining and provides the opportunity to learn about design elements from multiple sources.
The team worked on this learning algorithm to solve the tension between hardware manufactures – who seek to develop new systems – and designers, who prefer to make minimal changes to their web design work when new devices are launched. While designers would prefer to design for as few devices as possible, the incentive of hardware manufacturers is to offer a "new market."
The team's algorithm enabled a new kind of design-based machine learning, that can stream structured visual descriptors for page elements from a central repository. The algorithm also allows data to be collected and integrated with the repository for supervised learning applications like crowd-sourcing.
Scott Klemmer and Maxime Lim
View Project Summary
View Project Report
Read the Academic Paper
Read Stanford Report Announcement