Many clustering algorithms have been developed and researchers need to be able to compare their effectiveness. For some clustering problems, like web page clustering, different algorithms produce clusterings with different characteristics: coarse vs fine granularity, disjoint vs overlapping, flat vs hierarchical. The lack of a clustering evaluation method that can evaluate clusterings with different characteristics has led to incomparable research and results. QC4 solves this by providing a new structure for defining general ideal clusterings and new measurements for evaluating clusterings with different characteristics with respect to a general ideal clustering. The paper describes QC4 and evaluates it within the web clustering domain by comparison to existing evaluation measurements on synthetic test cases and on real world web page clustering tasks. The synthetic test cases show that only QC4 can cope correctly with overlapping clusters, hierarchical clusterings, and all the difficult boundary cases. In the real world tasks, which represent simple clustering situations, QC4 is mostly consistent with the existing measurements and makes better conclusions in some cases.

@INPROCEEDINGS{QC407,
    AUTHOR =    {Daniel Crabtree and Peter Andreae and Xiaoying Gao},
    TITLE =     {QC4 - A Clustering Evaluation Method},
    BOOKTITLE = {The 11th Pacific-Asia Conference on Knowledge Discovery 
                 and Data Mining (PAKDD'07)},
    YEAR =      {2007},
    PAGES =     {59--70}
}