Hybrid clustering based on content and connection structure using joint nonnegative matrix factorization

• Published in: Journal of global optimization 2019-08, Vol.74 (4), p.861-877
• Main Authors: Du, Rundong, Drake, Barry, Park, Haesun
• Format: Article
• Language: English
• Subjects: Algorithms; Clustering; Computer Science; Edge joints; Factorization; Materials handling; Mathematical analysis; Mathematics; Mathematics and Statistics; Matrix methods; Operations Research/Decision Theory; Optimization; Real Functions; Vectors (mathematics)
• ISSN: 0925-5001; 1573-2916
• DOI: 10.1007/s10898-017-0578-x

A hybrid method called JointNMF is presented which is applied to latent information discovery from data sets that contain both text content and connection structure information. The new method jointly optimizes an integrated objective function, which is a combination of two components: the Nonnegative Matrix Factorization (NMF) objective function for handling text content and the Symmetric NMF (SymNMF) objective function for handling network structure information. An effective algorithm for the joint NMF objective function is proposed so that the efficient method of block coordinate descent framework can be utilized. The proposed hybrid method simultaneously discovers content associations and related latent connections without any need for postprocessing of additional clustering. It is shown that the proposed method can also be applied when the text content is associated with hypergraph edges. An additional capability of the JointNMF is prediction of unknown network information which is illustrated using several real world problems such as citation recommendations of papers and leader detection in organizations. The proposed method can also be applied to general data expressed with both feature space vectors and pairwise similarities and can be extended to the case with multiple feature spaces or multiple similarity measures. Our experimental results illustrate multiple advantages of the proposed hybrid method when both content and connection structure information is available in the data for obtaining higher quality clustering results and discovery of new information such as unknown link prediction.