Incremental clustering of attributed graphs

An incremental clustering system based on a new criterion function to group patterns represented by attributed graphs is presented. The system takes a succession of attributed graphs and builds up a concept hierarchy that summarizes and organizes input instances incrementally. For this purpose, the authors propose a new criterion function based on entropy minimization, and present an incremental clustering algorithm with the criterion function. For the attributed graph as an input instance, the clustering algorithm incrementally obtains a concept hierarchy in a top-down manner using the hill climbing strategy. It is shown that the execution of the incremental clustering algorithm and classification algorithm can be interleaved since the concept hierarchy is constructed incrementally. Finally, the proposed method is applied to the clustering of simple examples to show its capability