IRM: integrated region matching for image retrieval
MULTIMEDIA '00 Proceedings of the eighth ACM international conference on Multimedia
Cluster ensembles --- a knowledge reuse framework for combining multiple partitions
The Journal of Machine Learning Research
| Hi-index | 0.00 |
Combining the results of different clustering to get a consensus clustering has attracted the attention of data mining researchers. In this context, it becomes necessary to measure diversity (or similarity) of a pair of partitions. Several diversity indices exist and these are based either on pairwise agreement or on clusterwise agreement. In pairwise agreement approach, similarity of two clusters is the number of common pairs of data elements. However, it is equally important to measure the level of disagreement rather than counting the frequency of disagreed pairs. We formulate this problem as a Transportation Problem and use Northwest Corner rule to compute feasible significance measures. We use this idea to propose a new index which differs from the existing measures in evaluating the extent of agreement by measuring the disagreement of data-pairs in terms of distance between cluster-pair of the disagreed data. We show experimentally that this yields a far better diversity index.