On Concise Representations of Frequent Patterns Admitting Negation

Abstract

The discovery of frequent patterns is one of the most important issues in the data mining area. An extensive research has been carried out for discovering positive patterns, however, very little has been offered for discovering patterns with negation. One of the main difficulties concerning frequent patterns with negation is huge amount of discovered patterns. It exceeds the number of frequent positive patterns by orders of magnitude. The problem can be significantly alleviated by applying concise representations that use generalized disjunctive rules to reason about frequent patterns, both with and without negation. In this paper, we examine three types of generalized disjunction free representations and derive the relationships between them. We also present two variants of algorithms for building such representations. The results obtained on a theoretical basis are verified experimentally.