Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
10325320 | Information Systems | 2005 | 26 Pages |
Abstract
This paper addresses some of the foundational issues associated with discovering the best few correlations from a database. Specifically, we consider the computational complexity of various definitions of the “top-k correlation problem,” where the goal is to discover the few sets of events whose co-occurrence exhibits the smallest degree of independence. Our results show that many rigorous definitions of correlation lead to intractable and strongly inapproximable problems. Proof of this inapproximability is significant, since similar problems studied by the computer science theory community have resisted such analysis. One goal of the paper (and for future research) is to develop alternative correlation metrics whose use will both allow efficient search and produce results that are satisfactory for users.
Related Topics
Physical Sciences and Engineering
Computer Science
Artificial Intelligence
Authors
Christopher Jermaine,