AI Seminar

Zoetrope & Ziggurat: Enhancing Interactions with the Dynamic Web

Eytan AdarAssistant ProfessorSchool of Information

The World Wide Web is generally treated by users and system designers as a single snapshot in time. In part, this limitation is due to the
tools which have been provided and to the data which is retained.
Browsers only provide a window to the "now Web," search engines generally store the most recent instance of a page, and those few Web sites that provide access to historical data do so through complex deep-web searches. The relationship between changing content and behavior has significant implications to search engines, Web sites, and end-user tools. I'll cover two such tools that break through the Now Web paradigm and help individuals and groups manipulate and utilize dynamic Web data.

The first, Zoetrope, is a system for working with the historical Web from the context of the more familiar Now Web. Zoetrope is based on a set of flexible operators that work with temporal content streams, allowing users to access, manipulate, and visualize temporal Web data
in novel ways. The power of Zoetrope is to automatically track and extract selected content in multiple versions of the same page (with current work on supporting this extraction between different pages).

Switching gears, I will also describe Ziggurat, a new tool for translation in Wikipedia that leverages the different rates of updates in different languages and communities (e.g., the French Wiki page for a French town will likely update census data more rapidly than the town's page in the English domain). Ziggurat is able to complete missing (semi-structured) information by learning the mappings between different languages without resorting to any explicit translation dictionary.

Both Zoetrope and Ziggurat demonstrate the benefits of considering the impact of time and change in the design of systems. Both represent an initial entry into a very rich space and I'll cover some ongoing and future work on both.
Eytan Adar is an Assistant Professor in the School of Information & Computer Science and Engineering at the University of Michigan. He
completed his doctoral work in the Computer Science and Engineering Department at the University of Washington. He works in the area of
temporal-informatics, studying how large populations interact with the dynamic Web and how those interactions can be enhanced. His interests are in understanding the dynamics of user behavior and data on the Web through text and log analysis, visualization, and the creation of new tools. Before entering graduate school, Eytan was a researcher at HP Labs and Xerox PARC for a number of years (spinning out a company called Outride somewhere in there). He received his Master of Engineering and Bachelor of Science degrees from the Massachusetts Institute of Technology. His website is at

Sponsored by