Graphical Data Mining

Abstract

Data mining has grown to include many more data types than the “traditional” flat files with numeric or categorical attributes. Images, text, video, and the internet are now areas of burgeoning data mining research. Graphical data is also an area of interest, since data in many domains—such as engineering design, network intrusion detection, fraud detection, criminology, document analysis, pharmacology, and biochemistry—can be represented in this form. Graph mining algorithms and methods are fewer and less mature than those designed for numerical or categorical data. In addition, the distinction between graph matching and graph mining is not always clear. In graph mining, we often want to find all possible frequent subgraphs of all possible sizes that occur a specified minimum number of times. That goal involves iteratively matching incrementally larger subgraphs, while classical graph matching is a single search for a static subgraph. Also, graph mining is an unsupervised learning task. Instead of searching for a single match to a specific graph, we are looking for known or unknown graphs embedded in the data.