Mining Frequent Subgraph Pattern Over a Collection of Attributed-Graphs and Construction of a Relation Hierarchy for Result Reporting

Graphs are compared to the standard attribute-value representations sophisticated data structures. Besides the description of an entity, a graph representation can also represent the relation of the entities to each other and by doing that it can be build up a complex network of knowledge pieces. These networks may be different kinds of networks such as telecommunication networks, computer networks, biological networks, and Web and social community networks. There are broad applications that require graph-based representations such as chemical informatics, bioinformatics, computer vision, video indexing, text retrieval, and Web analysis. Attributed graphs are a special form of graphs that describe the nodes and the edges of a graph by attributes. An important task of graph mining is mining frequent subgraph patterns. The summarization of graphs into groups of subgraphs are used for further characterization, discrimination, classification, and cluster analysis of a collection of graphs. In this paper, we introduce mining frequent subgraph pattern over a collection of attributed graphs. We describe the graph representation. We explain the fast graph-matching algorithm. How to deal with the similarity on the attributes on the nodes and the edges is described. Then, we explain the main part of our paper our algorithm for frequent subgraph mining. We describe how the found results can be reported in such a way that a human can easily overlook the results and how he can use the discovered knowledge for his application. The reporting is done by constructing a relation hierarchy over the discovered groups of subgraphs. The user gets the relation hierarchy of the subgraphs graphically displayed. We give results for our algorithm on an application of attributed image graphs. These image graphs have been obtained by automatic image analysis of ultra-sonic images of welding seams. The task was to classify the images into different defects such as pore, whole, and cracks.

[1]  Padraig Cunningham,et al.  Learning Feature Weights for CBR: Global versus Local , 1997, AI*IA.

[2]  Jiawei Han,et al.  Data Mining: Concepts and Techniques , 2000 .

[3]  Francesco Ricci,et al.  Structured Cases, Trees and Efficient Retrieval , 1998, EWCBR.

[4]  Douglas H. Fisher,et al.  Knowledge Acquisition Via Incremental Conceptual Clustering , 1987, Machine Learning.

[5]  Petra Perner,et al.  Using CBR Learning for the Low-Level and High-Level Unit of an Image Interpretation System , 1999 .

[6]  Ryszard S. Michalski,et al.  A Theory and Methodology of Inductive Learning , 1983, Artificial Intelligence.

[7]  Dong Yan,et al.  The Semantic Web for Complex Network Analysis in Biomedical Domain , 2015, 2015 7th International Conference on Information Technology in Medicine and Education (ITME).

[8]  Petra Perner.  Ultra Sonic Image Interpretation for Non-Destructive Testing , 1996, MVA.

[9]  P. Santhi Thilagam,et al.  Discovering suspicious behavior in multilayer social networks , 2017, Comput. Hum. Behav..

[10]  Petra Perner.  Different Learning Strategies in a Case-Based Reasoning System for Image Interpretation , 1998, EWCBR.

[11]  Petra Perner,et al.  Case-base maintenance by conceptual clustering of graphs , 2006, Eng. Appl. Artif. Intell..

[12]  Pat Langley,et al.  Models of Incremental Concept Formation , 1990, Artif. Intell..

[13]  Ana Paula Appel,et al.  Link and Graph Mining in the Big Data Era , 2017, Handbook of Big Data Technologies.

[14]  Horst Bunke,et al.  Similarity Measures for Structured Representations , 1993, EWCBR.

[15]  David W. Aha,et al.  A Review and Empirical Evaluation of Feature Weighting Methods for a Class of Lazy Learning Algorithms , 1997, Artificial Intelligence Review.