1Intelligent Database Systems Lab

國立雲林科技大學National Yunlin University of Science and Technology

SCALE: a scalable framework for efficiently clustering transactional data

Hua Yan · Keke Chen · Ling Liu · Zhang Yi

DMKD 2010

Reported by Wen-Chung Liao, 2010/03/02

2

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Outlines

Motivation Objective WCD clustering Evaluating clustering results Experiments Conclusions Comments

3

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Motivation

transactional data clustering algorithms require users to manually tune at least one or two parameters

lacks of cluster validation methods to evaluate the quality of transactional clustering results.

4

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Objectives

Present a fast, memory-saving, and scalable clustering algorithm that can efficiently handle large transactional datasets without resorting to manual parameter settings.

SCALE framework

5

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.WCD clustering

transactional dataset─ {abcd, bcd, ac, de, def}

6

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

7

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Evaluating clustering results

8

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Experiment

two synthetic datasets: ─ Tc30a6r1000_2L─ TxI4Dx Series

Three real datasets: ─ Zoo─ Mushroom ─ Retail

T10I4Dx

T10I4Dx

TxI4D100k

9

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

ZooTc30a6r1000_2L

10

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

11

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Conclusion

Two unique features of SCALE ─ the WCD clustering algorithm—a fast, memory-saving

and scalable method for clustering transactional data, ─ two transactional data specific cluster evaluation

measures: LISR and AMI. Some promising directions

─ perform some experimental comparison between the WCD measure and the entropy measure.

─ design a better algorithm for determining the best K for transactional data clustering.

─ Extend our work to handle transactional data streams

12

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Comments

Advantage─ No parameter setting required

Shortage─ If there is no BKPlot, WCD needs to determine K manually.─ No description of how BKPlot generates K in categorical c

ase. Applications

─ Transactions clustering─ Web log clustering─ …