Mining More Inter-Cluster Association Information to Improve Ensemble Clustering
Ruihan Li, Fusheng Yu · 2025
Ensemble clustering has gained increasing attention for its critical role in data analysis. Recent studies emphasize the importance of mining inter-cluster relationships instead of relying solely on pairwise similarities. However, existing methods often fail to fully exploit the global structure of inter-cluster association graphs, limiting their effectiveness. To address this, we propose a Deep Walk-based method to mine and utilize inter-cluster association information more effectively. By learning embeddings of cluster nodes, the approach captures multi-scale and global inter-cluster relationships, which are then incorporated into the co-association matrix to enhance clustering performance. Furthermore, we introduce an adaptive linkage selection strategy for hierarchical clustering as the consensus function. Unlike existing methods with fixed linkage modes, our approach leverages the inter-cluster association graph to adaptively select the most suitable linkage strategy, improving adaptability across diverse datasets. Experimental results on multiple datasets demonstrate that the proposed algorithm achieves superior performance and robustness, validating its effectiveness in ensemble clustering.