[HomePage] [Projects] [Papers] [Students]
This project investigates the applicability of perceptual visualization techniques for displaying data with no explicit spatial properties. Important "information visualization" environments of this type abound, for example, in areas like financial analysis and the exploration of web-based detests.
Our testbed application is a collection of datasets from the University of Michigan Trading Agent Competition. This system allows intelligent auction agents to compete in a simulated ecommerce auction environment. Each agent is tasked to buy certain items (e.g., a round-trip airline ticket, a hotel, and tickets to entertainment events like baseball games or the symphony). The agent is given additional information about the dates, prices, qualities, and quantities its owner prefers. The agents compete within a simulated environment containing multiple auctions running simultaneously for a fixed period of time.
Numerous options must be considered by each agent. The quantity of certain types of items (e.g., airline tickets to a given city on a particular day) may be fixed. Demand and supply can cause prices to fluctuate. Different auctions may be run using completely different rules (e.g., some auctions offer items until a fixed close date, while others end the auction after a specified period of inactivity). Some or all of these issues may need to be considered during the design of the auction agent.
A Trading Agent Competition (TAC) was recently run in July 2000 at the Fourth International Conference on Multiagent Systems (ICMAS-00). Results from the competition (including the affiliations and scores of the top eight agents) are available at the TAC homepage. Our software was used to visualize the results of each game. Feedback from the contestants was overwhelmingly positive. We will continue to improve and refine our tool, and plan to make it available to help TAC contestants design, analyze, and debug their agents. Perhaps most importantly, this software provides a critical proof-of-concept for the use of perceptual visualization techniques in non-spatial environments like ecommerce, web traces, and financial datasets.