Discovering Functional Dependencies and Association Rules by - - PowerPoint PPT Presentation

Discovering Functional Dependencies and Association Rules by Navigating in a Lattice of OLAP Views

Team LIS – IRISA (Rennes, France) Pierre Allard, pierre.allard@irisa.fr Sébastien Ferré, ferre@irisa.fr Olivier Ridoux, ridoux@irisa.fr 1

Introduction : photos database

Name Date Place Theme … T1 T2 …

• Rules extraction from multi-

valued set

▫ Set of described photos

• Traditional systems extract a

list of rules

• Many extracted rules
• Difficulty in finding

relevant information

2

Date  Place Date, Camera  Place (Date  Camera, Tp) Place = Marseille  Event = Conference …

Problem: Rules browsing

• Excessive number of rules

▫ Difficulty in finding relevant information

• Do not present all the rules together

▫ Show rules in subsets (views)

• Need interactivity

▫ Navigation ▫ Exploration

3

Rule extraction : background

• Multi-valued context

▫ Attr(r) = {Name, Date, Place, Theme, Event, Camera, …} ▫ Set of items described in each attribute

• Functional Dependencies

▫ Date  Place

• Conditional Functional Dependencies

▫ (Date  Camera, {Topic = Holiday})

• Association Rules

▫ (Place = Marseille)  (Event = Conference JIGOT)

• Hierarchy between rules with help of FCA [Medina09]

▫ Most general to most specific ▫ AR < CFD < FD

4

OLAP : background

• On-Line Analytical Processing

(OLAP) [Codd93]

• Business intelligence

(statisticians)

• Extract trends, charts from a

relation

• Cube: view of the relation

▫ Fixed dimensions and measure ▫ Aggregation ▫ Granularity levels ▫ Navigation from cube to cube

5

Projection of a relation into an OLAP cube

• Cube = aggregations of measures, by each value of

each dimension

• Dimensions  measure

▫ Dim(c) = (Orientation, Date), Meas(c) = Camera

• Premises  conclusion

▫ Orientation, Date  Camera ?

• Classic cube with no aggregation

▫ Multisets at each cell ▫ Keep all the values of the measure

 Support, confidence

• Subset of rules, regarding dimensions and measure

6

View : Photo database

7

Name Date Orientation Camera t_1 DSC01 16 jul Landscape Nikon t_2 DSC02 16 jul Landscape Nikon t_3 DSC03 16 jul Portrait Nikon t_4 DSC04 21 aug Portrait Nikon t_5 IMG05 21 aug Portrait Apple Projection Dim(c) = (Orientation, Date) Meas(c) = Camera (Orient, Date) Landscape Portrait 16 jul {{Nikon, Nikon}} {{Nikon}} 21 aug {{Nikon, Apple}}

View : Rules in projection

• Cell with a unique item value

▫ Association Rule (Orientation = Landscape), (Date = 16 jul)  (Camera = Nikon)

• Subset of cube with empty or unique item values cells

▫ Conditional Functional Dependency (Orientation, Date  Camera, {Orientation = Landscape})

• Cube with empty or unique item values cells

▫ Functional Dependency

8

(Orient, Date) Landscape Portrait 16 jul {{Nikon, Nikon}} {{Nikon}} 21 aug {{Nikon, Apple}}

View : Rules in projection

• Respected hierarchy regarding number of cells

▫ AR < CFD < FD

• Display of all rules type in one view
• Accessible support et confidence
• Must navigation to access other cubes

9

View : Granularity

• Dimension values : taxonomy organized in levels
• New possibilities of cube creation, regarding

new dimensions and measures

▫ e.g. datemonth, dateyear, placecountry

10

2010/01/09 2010/01/16 2010/02/06 2010/02/07 2010/01 2010/02 2010 dateday datemonth dateyear

View : Granularity

• Date  Place
• Datemonth  Place
• Add new relevant information
• Need navigation links, to go from a cube to

another

11

16 jul 09 17 jul 09 28 jan 10 {{Marseille}} {{Marseille}} {{Hammamet}} jul 09 jan 10 {{Marseille, Marseille}} {{Hammamet}}

Navigation

• Use of standard OLAP navigation links

▫ Add / Delete a dimensions ▫ Drill-down / Roll-up a dimension

 Delete is a specific case of roll-up

• Introduction of new navigation links

▫ Drill-down / Roll-up the measure ▫ Change measure

12

Navigation : predictable consequences

• Rules appearance
• Rules disappearance
• Rules preservation

13

 Unic item value   Multiple item values  Cube with FD  Cube without FD

Abilis : Prototype

• Based on Logical Information System (LIS)

▫ Camelis kernel with web interface [Ferre09] ▫ Query and navigate context with logical formulas

• Current query

▫ Allows you to select a subset of items (conjonctions, disjonctions, negations, etc)

• Navigation tree

▫ Summary of the dimensions values according to the current selection

• OLAP navigation links

▫ Granularity created by LIS

14

15

Initial view : query, navigation tree and extension

16

Navigation tree = index

17

Add « Theme » partition

18

Create complex selections

19

Add « event » as measure

20

Set « date » by day as partition : FD

21

Roll-up date by day to date by month : CFD

22

Add « Camera model » as partition : CFD

23

Remove birthday events : FD

Conclusion and perspectives

• Show FD, CFD and AR in one view
• Predict some consequences on rules with

navigation links

• Manage more complex data (relational, with

zero or multiple attributes)

• Add indicator values for navigation links
• Allow user to create cubes from aggregated data

24

Thank your for your attention

25 Team LIS – IRISA (Rennes, France) Pierre Allard, pierre.allard@irisa.fr Sébastien Ferré, ferre@irisa.fr Olivier Ridoux, ridoux@irisa.fr