Methods for Linking and Mining Massive Heterogeneous Databases

Manyreal-world KDDexpeditions involve investigation of relationships betweenvariables in different, heterogeneousdatabases. Wepresent a dynamic programmingtechnique for linking records in multiple heterogeneousdatabases usinglooselydefinedfields that allowfree-style verbatim entries. Wedevelop an interestingness measurebased on non-parametric randomization tests, whichcan be used for miningpotentially useful relationships amongvariables. This measure usesdistributional characteristics of historical events, hence accommodatingvariable-length records in a natural way.Asan illustration, we include a successful application of the proposed methodologyto a real-world data miningproblem at LucentTechnologies.

• Type: Conference
• Year: 1998
• Where: KDD
• Authors: José C. Pinheiro, Don X. Sun
• Keywords: Allowfree-style Verbatim Entries, Data Mining, KDD 1998, KDDexpeditions Involve Investigation, Non-parametric Randomization Tests