Reliable Co-Occurrence Guided Partial Clustering Ensemble.

Abstract

Partial clustering ensemble integrates multiple incomplete base clusterings to generate a consensus one. The key challenge of partial clustering ensemble is to infer the consensus clustering from the incomplete and unreliable base clusterings without access to the original data. To this end, we propose a novel partial clustering ensemble model that first extracts two types of highly reliable information from initial base results. Then, the proposed model constructs a similarity matrix with the guidance of the extracted highly reliable information, the available incomplete ensemble base clusterings, and the priors of a valid similarity matrix. We solve the proposed model using the alternating direction method of multipliers (ADMMs) algorithm. Extensive experimental comparisons demonstrate that our approach significantly improves clustering performance on incomplete base clusterings, outperforming state-of-the-art methods by a large margin (e.g., 19% increase in normalized mutual information (NMI) scores). The code of this article is available at: https://github.com/qt379/Reliable-Co-occurrence-Guided-Partial-Clustering-Ensemble.