From Context to Distance-Learning Dissimilarity for Categorical Data Clustering

1. From Context to Distance-Learning Dissimilarity for Categorical Data Clustering Presenter : Jian-Ren ChenAuthors : DINO IENCO, RUGGERO G. PENSA, and ROSA MEO 2012 , ACMKDD

2. Outlines • Motivation • Objectives • Methodology • Experiments • Conclusions • Comments

3. Motivation • Clustering data described by categorical attributesis a challenging task in data mining applications. • It is difficult to define a distance between pairs of values of a categorical attribute, since the values are not ordered.

4. Objectives • We present a new methodology to compute a context-based distancebetween values of a categorical variable. • - apply this technique to hierarchical clusteringof categorical data.

5. Methodology-Framework • DILCA (DIstance Learning for Categorical Attributes) • selection of a suitable context: • a parametric method • a fully automatic one • compute the distance between any pair of values of a specific categorical attribute

6. Methodology - Context Selection

7. Methodology- Context Selection

8. Methodology - Context Selection

9. Methodology -Distance Computation

10. Experiments - Datasets

11. Experiments-Purity、NMI、ARI

12. Experiments -Purity、NMI、ARI

13. Experiments-Purity、NMI、ARI

14. Experiments -Impact of σ on DILCAM

15. Experiments -Impact of σ on DILCAM

16. Experiments-Scalability

17. Conclusions • DILCA is competitive with respect to the stateof the art of categorical data clustering approaches. • DILCA is scalable and has alow impact on the overall computational time of a clustering task.

18. Comments • Advantages • scalable, computational time • Applications • a context-based distance between values of a categorical variable • hierarchical clustering of categorical data