Evaluation of Community Mining Algorithms in the Presence of Attributes

Grouping data points is one of the fundamental tasks in data mining, commonly known as clustering. In the case of interrelated data, when data is represented in the form of nodes and their relationships, the grouping is referred to as community. A community is often defined based on the connectivity of nodes rather than their attributes or features. The variety of definitions and methods and its subjective nature, makes the evaluation of community mining methods non-trivial. In this paper we point out the critical issues in the common evaluation practices, and discuss the alternatives. In particular, we focus on the common practice of using attributes as the ground-truth communities in large real networks. We suggest to treat these attributes as another source of information, and to use them to refine the communities and tune parameters.

• Type: Journal
• Year: 2015
• Where: PAKDD
• Authors: Reihaneh Rabbany, Osmar R. Zaïane
• Keywords: Data Mining, PAKDD 2015