QED: An Efficient Framework for Temporal Region Query Processing

1
QED An Efficient Framework for Temporal Region
Query Processing

• Yi-Hong Chu ???
• Network Database Laboratory
• Dept. of Electrical Engineering
• National Taiwan University

2
Introduction

• Dense Region Query
• Data records are viewed as data points in the
  d-dimensional data space constructed by the
  d-attributes.
• Locate the regions with higher density than their
  surroundings.

Salary (1000)
Dense region
Age
3
Grid-based Approach

• The data space is divided into non-overlapping
  rectangular grids (cells).
• Density of a cell
• the percentage of data points contained in this
  cell

Salary (1000)
Dense cell
Maximal connected dense cells
Dense region
Age
0 10 20 30 40 50 60 70 80 90 100
4
Motivation

• Previous research tends to ignore the time
  feature of the data.
• They execute queries over the entire database.
• However, different dense regions may be
  discovered if different time periods are taken
  into consideration.
• (the density of a cell )
• Discovering dense regions over different time
  intervals is crucial for users to get the
  interesting patterns hidden in data.

5
Example

• Some dense regions may exist in certain time
  intervals but will not be discovered if taking
  all data records into account.
• Middle-aged people

ltAgt the number of customers in
different time slots ltBgt the number of
middle-aged people in different time
slots
6
Temporal Dense Region Query

• Dense Region Discovery in the constrained time
  intervals.
• E.g., each Sunday in May,
• Time slots
• Derived by segmenting the data points with a time
  granularity, e.g. hour, week, month, etc.
• For users to specify a variety of time periods of
  interest
• Problem Definition
• Given a set of time slots, and the density
  threshold ?, find the dense regions in the
  queried time slots.

7
QED Framework

• Challenge
• The queried time intervals are unknown in
  advance.
• QED (Querying tEmporal Dense region)
• Offline Maintaining Phase
• Construct a summarized data structure, RF-tree,
  for each time slot
• Online Clustering Phase
• Answer various user queries based on the RF-trees

8
QED Framework

9
Offline Maintaining Phase- Construct the RF-trees

• Basic Idea
• A number of cells having nearly the density value
  can be summarized by their average density value.
• Uniform Region
• A region where the cells contained in it have
  nearly of the same density value

region
10
Uniform Region

• Entropy-based approach
• Entropy of a region
• Maximum entropy of a region
• Uniform region

11
Example (Uniform Region)

• Case 1
• Case 2

Region A
Region A
12
Construct the RF-tree

• Recursively partition the data space to find the
  uniform region
• The leaf nodes will be of two cases
• A cell
• A uniform region
• RF (Region Feature)

13
Online Query Processing Phase

• Step1
• Combine the RF-trees of the queried time slots.
• Step2
• Execute the query on the combined RF-tree.

14
Step1 Combine the RF-trees

• Three cases for combining the corresponding
  regions in two RF-trees.
• Case 1 Both are uniform regions
• Case 2 Both are non-uniform regions
• Case 3 Only one is a uniform region

15
Step2 Execute the query

• All leaf nodes in the combined RF-trees are
  examined to discover the dense cells in the data
  space.
• The leaf nodes will be of two cases
• A cell
• A uniform region compare the average density
  with the density threshold?
• The leaf nodes containing dense cells will be put
  into a queue for further dense region discovery.

16
Conclusion

• The problem of temporal dense region query is
  explored to discover dense regions in the queried
  time slots.
• We also propose the QED framework to execute
  temporal dense region queries.
• QED is advantageous in that various queries with
  different density thresholds and time slots can
  be efficiently supported by using the concept of
  time slot and proposed RF-tree.

17
References

• Yi-Hong Chu, Kun-Ta Chuang, Ming-Syan Chen, QED
  an Efficient Framework for Temporal Dense Region
  Processing, in Proc. of PAKDD, 2005.
• W. Wang, J. Yang, and R. Muntz1997, STING A
  Statistical Information Grid Approach to Spatial
  Data Mining, in Proc. of VLDB, 1997.
• D,-S. Cho, B-H.Hong, and J.Max. Efficient Region
  Query Processing by Optimal Page Ordering. In
  Proc. of ADBIS-DASFAA, 2000.

18
Thank You

• Q A