Mining Association Rules in Graphs Based on Frequent Cohesive Itemsets

• Published in: Advances in Knowledge Discovery and Data Mining pp. 637 - 648
• Main Authors: Hendrickx, Tayena, Cule, Boris, Meysman, Pieter, Naulaerts, Stefan, Laukens, Kris, Goethals, Bart
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing 2015
• Series: Lecture Notes in Computer Science
• Subjects: Association Rule; Frequent Subgraph; Mining Association Rule; Node Label; Nucleic Acid Research
• ISBN: 3319180312; 9783319180311
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-18032-8_50

Searching for patterns in graphs is an active field of data mining. In this context, most work has gone into discovering subgraph patterns, where the task is to find strictly defined frequently re-occurring structures, i.e., node labels always interconnected in the same way. Recently, efforts have been made to relax these strict demands, and to simply look for node labels that frequently occur near each other. In this setting, we propose to mine association rules between such node labels, thus discovering additional information about correlations and interactions between node labels. We present an algorithm that discovers rules that allow us to claim that if a set of labels is encountered in a graph, there is a high probability that some other set of labels can be found nearby. Experiments confirm that our algorithm efficiently finds valuable rules that existing methods fail to discover.