Seminars & Events

Mathematics and Computer Science Division
"Data Mining Techniques for the Data Warehouse"

DATE: June 22, 2007
TIME: 10:30 am
SPEAKER: Svetlozar Nestorov, University of Chicago
LOCATION: Building 221 - A216, Argonne National Laboratory
HOST: Ian Foster

Description:
As the amount of data generated by business applications and scientific experiments continues to grow, so does the importance of capturing, storing, and ultimately mining this data. While storage organization and structure can vary dramatically, the most popular choice is a data warehouse build on top of a relational database system.

In my talk, I will present several techniques for discovering patterns and associations in relational databases, applicable to data warehouses with varying structure such normalized and star schemas. I will describe the query-flocks framework which enables the declarative formulation of data mining queries over any database schema as well as their efficient implementation. I will also present qualified association rules (QAR) which are designed for mining data warehouses with the star schema and discover associations among attributes of multiple relations.

Sum constraints queries are another common type of query involving attributes of multiple relations. I will describe optimization techniques for such queries based on partial constraint derivation and query rewriting. I will also present a ranking framework for efficiently generating any K (partial) results of sum constraints queries.

Save the event to your calendar [schedule.ics]