Post-processing Association Rules: A Network Based Label Propagation Approach

Abstract

Association rules are widely used to find relations among items in a given database. However, the amount of generated rules is too large to be manually explored. Traditionally, this task is done by post-processing approaches that explore and direct the user to the interesting rules. Recently, the user’s knowledge has been considered to post-process the rules, directing the exploration to the knowledge he considers interesting. However, sometimes the user wants to explore the rule set without adding his prior knowledge BIAS, exploring the rule set according to its features. Aiming to solve this problem, this paper presents an approach, named $PAR_{LP}$ (Post-processing Association Rules using Label Propagation), that explores the entire rule set, suggesting rules to be classified by the user as “Interesting” or “Non-Interesting”. In this way, the user is directed to analyze the rules that have some importance on the rule set, so the user does not need to explore the entire rule set. Moreover, the user’s classification is propagated to all the rules using label propagation approaches, so the most similar rules will likely be on the same class. The results show that the $PAR_{LP}$ succeeds to direct the exploration to a set of rules considered interesting, reducing the amount of association rules to be explored.