Transfer Clustering Ensemble Selection
Résumé
Clustering ensemble (CE) takes multiple clusteringsolutions into consideration in order to effectively improve theaccuracy and robustness of the final result. To reduce redundancyas well as noise, a CE selection (CES) step is added to furtherenhance performance. Quality and diversity are two importantmetrics of CES. However, most of the CES strategies adoptheuristic selection methods or a threshold parameter setting toachieve tradeoff between quality and diversity. In this paper, wepropose a transfer CES (TCES) algorithm which makes use of therelationship between quality and diversity in a source dataset, andtransfers it into a target dataset based on three objective functions.Furthermore, a multiobjective self-evolutionary process isdesigned to optimize these three objective functions. Finally, weconstruct a transfer CE framework (TCE-TCES) based on TCESto obtain better clustering results. The experimental results on 12transfer clustering tasks obtained from the 20newsgroups datasetshow that TCE-TCES can find a better tradeoff between qualityand diversity, as well as obtaining more desirable clusteringresults.