Abstract | ||
---|---|---|
We examine whether the quality of different clustering algorithmscan be compared by a general, scientifically sound procedure whichis independent of particular clustering algorithms. We argue thatthe major obstacle is the difficulty in evaluating a clusteringalgorithm without taking into account the context: why does the usercluster his data in the first place, and what does he want to dowith the clustering afterwards? We argue that clustering should notbe treated as an application-independent mathematical problem, butshould always be studied in the context of its end-use. Differenttechniques to evaluate clustering algorithms have to be developedfor different uses of clustering. To simplify this procedure weargue that it will be useful to build a “taxonomy of clusteringproblems” to identify clustering applications which can be treatedin a unified way and that such an effort will be more fruitful thanattempting the impossible – developing “optimal” domain-independentclustering algorithms or even classifying clustering algorithms interms of how they work. |
Year | Venue | Field |
---|---|---|
2012 | ICML Unsupervised and Transfer Learning | Computer science,Artificial intelligence,Cluster analysis,Machine learning |
DocType | Citations | PageRank |
Journal | 4 | 0.41 |
References | Authors | |
0 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
von luxburg | 1 | 3246 | 170.11 |
Robert C. Williamson | 2 | 4191 | 755.22 |
Isabelle Guyon | 3 | 11033 | 1544.34 |