Inferring Knowledge from Frequent Patterns

Abstract

Many knowledge discovery problems can be solved efficiently by means of frequent patterns present in the database. Frequent patterns are useful in the discovery of association rules, episode rules, sequential patterns and clusters. Nevertheless, there are cases when a user is not allowed to access the database and can deal only with a provided fraction of knowledge. Still, the user hopes to find new interesting relationships. In the paper, we offer a new method of inferring new knowledge from the provided fraction of patterns. Two new operators of shrinking and extending patterns are introduced. Surprisingly, a small number of patterns can be considerably extended into the knowledge base. Pieces of the new knowledge can be either exact or approximate. In the paper, we introduce a concise lossless representation of the given and derivable patterns. The introduced representation is exact regardless the character of the derivable patterns it represents. We show that the discovery process can be carried out mainly as an iterative transformation of the patterns representation.