|
|
MIT Electrical Engineering and Computer Science
EECS Event |
Monday, April 9, 2001
4:15 PM (refreshments 4:00)
Room NE43-518
EECS Special Seminar
Abstract
Information extraction from large datasets is a complex, iterative process, often involving long frustrating delays, with little feedback to the user. This talk is about interactive query processing, and focuses on techniques to provide continual incremental results, and support dynamic user control of the processing.
I first present an Online Reordering mechanism to support dynamic user control of processing for long-running queries. I then argue that traditional query processing architectures are too rigid for providing continual results, and describe an architecture for providing incremental query processing results at a fine granularity, in the form of incomplete output tuples with missing attributes. This architecture, coupled with online reordering, is used in the Telegraph adaptive dataflow system for querying diverse data sources on the Internet. User actions at the Telegraph user interface are translated into preferences for different parts of the result. These actions are fed back into the Telegraph system, which accordingly adapts its internal dataflow to track the user preferences.