Interactive Mining of Correlations - A Constraint Perspective

The problem of nding minimal sets of objects, that are correlated and have statistically signiicant number of occurrences in a database, has recently received considerable attention, as an alternative to association rules. A case for making mining human-centered, interactive, and exploratory via application speciic constraints was recently made in the papers 14, 12]. The technical results of these papers were established in the context of eecient computation of frequent sets satisfying user speciied constraints. In this preliminary report, we examine the problem of eecient computation of correlated sets satisfying given constraints. We discuss three algorithms for this purpose and report on the results of their experimental evaluation.