TY - GEN
T1 - Structured Bipartite Graph Ensemble Clustering
AU - Wang, Chen
AU - Hou, Feng
AU - Wang, Yi
AU - Wang, Ruili
N1 - Publisher Copyright:
© 2024 Copyright held by the owner/author(s). Publication rights licensed to ACM.
PY - 2024/12/28
Y1 - 2024/12/28
N2 - Conventional bipartite graph-based cluster ensembles usually compute a weight matrix from a base clustering result to represent the similarities between the samples and clusters. These weight matrices are concatenated or averaged as a bipartite weight matrix with which a bipartite graph is created, and then graph-based partition techniques are used to partition the samples in the bipartite graph as the final clustering result. However, such methods may suffer from unreliable clusterings and it is difficult to find an explicit cluster structure from the bipartite graph since the clusters in the base clusterings are usually diverse. In this paper, we propose a novel Structured Bipartite Graph Ensemble Learning (SBGEC) for cluster ensembles. To address these issues, SBGEC constructs a sample-cluster similarity matrix for each base clustering and introduces a weight matrix rearrangement technique to optimize cluster correspondences. Multiple bipartite graphs are generated using these matrices, and a structured bipartite graph is learned to combine the original graphs while maintaining a clear cluster structure. Experimental results on synthetic and real-world datasets demonstrate the superior performance of SBGEC compared to state-of-the-art clustering techniques.
AB - Conventional bipartite graph-based cluster ensembles usually compute a weight matrix from a base clustering result to represent the similarities between the samples and clusters. These weight matrices are concatenated or averaged as a bipartite weight matrix with which a bipartite graph is created, and then graph-based partition techniques are used to partition the samples in the bipartite graph as the final clustering result. However, such methods may suffer from unreliable clusterings and it is difficult to find an explicit cluster structure from the bipartite graph since the clusters in the base clusterings are usually diverse. In this paper, we propose a novel Structured Bipartite Graph Ensemble Learning (SBGEC) for cluster ensembles. To address these issues, SBGEC constructs a sample-cluster similarity matrix for each base clustering and introduces a weight matrix rearrangement technique to optimize cluster correspondences. Multiple bipartite graphs are generated using these matrices, and a structured bipartite graph is learned to combine the original graphs while maintaining a clear cluster structure. Experimental results on synthetic and real-world datasets demonstrate the superior performance of SBGEC compared to state-of-the-art clustering techniques.
KW - Clustering
KW - Ensemble Clustering
KW - Structure Learning
UR - http://www.scopus.com/inward/record.url?scp=85216200679&partnerID=8YFLogxK
U2 - 10.1145/3696409.3700265
DO - 10.1145/3696409.3700265
M3 - Conference contribution
AN - SCOPUS:85216200679
T3 - Proceedings of the 6th ACM International Conference on Multimedia in Asia, MMAsia 2024
BT - Proceedings of the 6th ACM International Conference on Multimedia in Asia, MMAsia 2024
PB - Association for Computing Machinery, Inc
T2 - 6th ACM International Conference on Multimedia in Asia, MMAsia 2024
Y2 - 3 December 2024 through 6 December 2024
ER -