Numerous clustering algorithms were proposed in the past years. However, the following two key problems have not been sufficiently addressed:

  1. Clustering, validating, and refining irregularly shaped (non-spherical) clusters.
  2. Incorporating the domain knowledge into the clustering process.

Automated algorithms are not very effective in clustering irregularly shaped clusters or incorporating the domain knowledge into clustering. Traditional statistical validation methods also do not work for validating the irregularly shaped clusters.-- Are there any effective methods handling these particular clustering structures? We propose an interactive visualization based approach and develop the VISTA visual cluster rendering system for validating and refining irregularly shaped clusters. With VISTA system it is also very convenient to incorporate any available domain knowledge into the clustering process.


  Visual Cluster Rendering System - validating and refining irregular clusters via interactive visualization
Authors: faculty Ling Liu
              student Keke Chen, at DISL
1. iVIBRATE System Framework for Large Datasets and UI of VISTA cluster rendering system
2. VISTA Interactive Techniques
3. Some heuristic rendering Rules 
4. Visualization of several datasets
5. Exploring web log data 1, 2.
6. Related publications


  free web counter