automated conceptual abstraction of large diagrams by daniel levy and christina christodoulakis...

Automated Conceptual Abstraction of Large Diagrams

By Daniel Levy and Christina ChristodoulakisDecember 2012

(2 days before the end of the world)

Introduction Big picture Clustering Algorithm Experiment & Results Conclusion

Outline

Introduction Big picture Clustering Algorithm Experiments & Results Conclusion

Outline

So what is this “clustering” you speak of? Why do we need to cluster? Reduce cognitive load

Introduction

Introduction

Big picture Clustering Algorithm Experiment + Results Conclusion

Outline

Big Picture

Vision

Diagram Abstraction

Its been done before..

Related Works

Consider a diagram stripped of semantics, or pre processed using methodologies in previous work

Cluster graph

Evaluate clusters proposed based on closeness of meaning in the node names

Our Approach

Our Approach

Introduction Big picture

Clustering Algorithm Experiment + Results Conclusion

Outline

Min-Cut

Naïve Min-Cut Algorithm

C

A

N B

1

2

3C

A

N B2

3

E4

E4

*Must result in exactly 2 partitions

Combinations / Creating partitions

*Assume there exist additional nodes

C

A

N B

1

2

3C

A

NB

1

E E4 4

C

D

A

B

21

3C

D

A

B

2

Minimum sets

C

D

A

B

21

3 C

D

A

B

2

3

D

AB

1

3

2D

AB3

2

D

AB

1

3

2D

AB

2

Cycles

E

D

C

A

B

12

3

4

5

Listing the min-cuts

E

D

C

A

B

12

3

4

5

Listing the min-cuts

E

D

C

A

B

12

3

4

Listing the min-cuts

5

E

D

C

A

B

12

3

4

5

Listing the min-cuts

E

D

C

A

B

12

3

4

5

Listing the min-cuts

E

D

C

A

B

12

3

4

5

E

D

C

A

B

12

3

Outside-in approach

E

D

C

A

B

12

3

4

5

E

D

C

A

B

12

35

Outside-in approach

E

D

C

A

B

12

3

4

5

E

D

C

A

B

12

3

4

E

D

C

A

B

12

3

4

5

We use RiTa WordNet getDistance() function We calculate pairwise distances between

nodes. Select for each node the smallest distance

between it and another node Sum all minimum distances Average over all nodes in candidate cluster

Cluster Distance Measure

Introduction Big picture Clustering Algorithm

Experiments + Results Conclusion

Outline

Experiment 1

Experiment #1

Experiment # 1User 1 abstraction

ExperimentationUser 2 abstraction

Experiment # 1automated abstraction

Experiment 2

Experiment #2

Simplified version

Introduction Big picture Clustering Algorithm Experiments + Results

Conclusion

Outline

Surprised at how similar manual clustering and automated clustering were.

Suggested improvements: Automatic distance threshold Creating subgraphs Strictness of clustering (min # of clusters Advanced min-cut discovery

Conclusions

Questions?Merry Christmas!