Tracking High Quality Clusters over Uncertain Data Streams

1
Tracking High Quality Clusters over Uncertain
Data Streams

• Chen Zhang, Ming Gao and Aoying Zhou
• ICDE 2009

2
Outline

• Introduction
• Uncertain Stream Model
• Definition
• Clustering Algorithm
• Experiment
• Conclusion
• My thought

3
Introduction

• Goal of clustering
• To find some clusters in one dataset.
• Cluster over uncertain data
• Data quality should be considered
• For example

4
Uncertain Stream Model

• Probability density function (pdf)
• Denote as (Dt, ft)
• Discrete probability
• In order to approximate the pdf
• Denote as (xt1, 0.2) (xt2, 0.3)(xtk, 0.35)
• Standard error/deviation
• Denote as xt r

5
Definition

• Definition 1 (instance uncertainty)
• U(xti) -log2(pt(xti))
• Definition 2 (tuple uncertainty)
• For example
• U(xa) -6 (1/6) log2(1/6) 2.585
• U(xb) -2 (1/2) log2(1/2) 1

6
Clustering Algorithm
7
Implementation

• Two phase method
• Reduce computation
• Firstly (K-NN)
• Select top k clusters
• Second
• ?U(C) U(C) U(C U xt, ut)

8
Experiment

• Validation Measures
• UM

9
(No Transcript)
10
Conclusion

• This paper main contributions
• Propose a two phase clustering over uncertain
  data stream
• Add quantify tuple uncertainty

11
My thought

• The idea is very simple.
• How to implement on sliding window model?