User-Guided Large Attributed Graph Clustering with Multiple Sparse Annotations

Abstract

One of the key challenges in large attributed graph clustering is how to select representative attributes. Previous studies introduce user-guided clustering methods by letting a user select samples based on his/her knowledge. However, due to knowledge limitation, a single user may only pick out the samples that s/he is familiar with while ignore the others, such that the selected samples are often biased. We propose a framework to address this issue which allows multiple individuals to select samples for a specific clustering. With wider knowledge coming from multiple users, the selected samples can be more relevant to the target cluster. The challenges of this study are two-folds. Firstly, as user selected samples are usually sparse and the graph can be large, it is non-trivial to effectively combine the different annotations given by the multiple users. Secondly, it is also difficult to design a scalable approach to cluster large graphs with millions of nodes. We propose the approach CGMA (Clustering Graphs with Multiple Annotations) to address these challenges. CGMA is able to combine the crowd’s consensus opinions in an unbiased way, and conducts an effective clustering with low time complexity. We show the effectiveness and efficiency of the proposed approach on real-world graphs, by comparing with existing attributed graph clustering approaches.