The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) CoDiNMF: Co-Clustering of Directed Graphs via NMF Woosang Lim Rundong Du Haesun Park School of Computational Science School of Computational Science School of Computational Science and Engineering and Engineering and Engineering Georgia Institute of Technology, Georgia Institute of Technology, Georgia Institute of Technology, Atlanta, GA 30332, USA Atlanta, GA 30332, USA Atlanta, GA 30332, USA
[email protected] [email protected] [email protected] Abstract co-clustering (Dhillon, Mallela, and Modha 2003), multi- view co-clustering (Sun et al. 2015), co-clustering based Co-clustering computes clusters of data items and the related on spectral approaches (Wu, Benson, and Gleich 2016; features concurrently, and it has been used in many appli- Rohe, Qin, and Yu 2016), co-clustering based on NMF cations such as community detection, product recommenda- tion, computer vision, and pricing optimization. In this paper, (Long, Zhang, and Yu 2005; Wang et al. 2011). However, we propose a new co-clustering method, called CoDiNMF, most of co-clustering methods assume that the connections which improves the clustering quality and finds directional between entities are symmetric or undirected, but many in- patterns among co-clusters by using multiple directed and teractions in real networks are asymmetric or directed. For undirected graphs. We design the objective function of co- example, the data set of patents and words contains a cita- clustering by using min-cut criterion combined with an addi- tion network among patents, and it can be represented as tional term which controls the sum of net directional flow be- a directed and asymmetric graph.