Title:    Extending Dependencies with Conditions to Capture
          Data Inconsistencies

Speaker:  Wenfei Fan
          LFCS, School of Informatics, University of Edinburgh &
          Bell Labs, Actel-Lucent

Abstract:
Real-world data is often dirty. A central problem for data cleaning
is how to characterize the consistency of the data, i.e., how to tell
whether the data is clean or not. Previous work on data cleaning
typically approaches this based on functional and inclusion
dependencies: inconsistencies in the data are detected as violations
of the dependencies. However, these dependencies often fail to
capture severe yet common inconsistencies in the data. This is not
surprising: traditional dependencies were developed for schema design
rather than data cleaning.

In light of this I shall introduce an extension of functional and
inclusion dependencies with conditions, which assure bindings of
semantically related data values. These conditional dependencies are
capable of detecting inconsistencies beyond what their traditional
counterpart can find, and in addition, are important in contextual
schema matching and data integration. I shall present fundamental
results for conditional dependencies including their consistency,
implication and inference analyses. Furthermore, I shall show how
these conditional dependencies can be used to improve data quality,
highlighting open research issues.

(joint work with Philip Bohannon, Yahoo Research!, and Floris
Geerts, Anastasios Kementsietsidis, Loreto Bravo, Xibei Jia,
and Shuai Ma, University of Edinburgh)