Visualization of Barter Trade

1
Visualization of Barter Trade

• By Dai Hanbo

2
Barter Trade System

• Barter Trade exchange
• Act as broker in the barter pool
• Member companies
• Using private label currency for transactions
• Cash flow feature
• Transaction cycles
• Cycle facilitate trade
• Transactions are optimized as cycles in the long
  run

3
Objectives

• Decompose the transactions into a set of cycles
  of cash flow
• Measuring the brittleness
• Detecting company communities
• Graphically display the results of these
  algorithms and the analyses of the networks

4
Cycle-decomposing

• Goldberg-Trajans Minimum-cost Circulation
  Algorithms can solve this in a general graph
• Graph not necessary balanced
• Cycles found have the Minimum Cycle Mean, which
  have no significant meaning in barter context.
• Three Decomposing Algorithms based on
  repetitively canceling particular cycles
• Arbitrary decomposition
• Greedy decomposition
• Profitable decomposition

5
Finding cycle Algorithms

• Arbitrary finding cycles
• Arbitrarily find any edge
• Greedy finding cycles
• Greedily find the edge with bigger flow
• Profitable cycles
• Take an unfolding style
• More profitable cycles with different lengths

6
Sample Transaction Network
7
Node
0 1 2 3
Level k
0 1 2 3 4
4
2
4
2
4
2
3
3
2
6
4
2
3
6
2
F value (given edge flow )
M value (maximum flow capacity )
8
Sample Transaction network with 10 nodes
9
Results of Profitable cycle-finding Algorithm

• Algorithm stars from node 5.
• Find more profitable cycles with different
  lengths
• cycle 5-gt7-gt4-gt2-gt5 (4 1) is not shown because
  of cycle 5-gt3-gt0-gt2-gt5 (45)

10
Results of three decompose algorithms
11
Measuring brittleness

• Brittleness represents importance of a company or
  a transact
• Define the importance by the revenue loss if a
  company leaves or a transaction gets canceled.
• Finding the shortest cycle to delete in order to
  minimize the loss
• Shortest Cycle Algorithm using the unfolding idea
  to find a shortest cycle

12
Importance of a company

• Company 0 is leaving the barter pool
• 4 cycles got canceled
• Revenue loss 26

13
Importance of a transaction

• Transaction from company 1 to company 7 is
  deleted
• Two cycles are affected
• Total loss 33

14
Detecting company communities

• Some companies may conduct more transactions
  inside a community than with companies in other
  communities
• Related Algorithms
• Spectral Bisection and Kernighan-Lin
• Hierarchical clustering
• Edge-betweenness

15
Edge-betweenness Algorithm

• Newmans Edge-betweenness Algorithm is used to
  detect these communities by iteratively canceling
  edges with the highest rank of betweenness
• A new dataset is made deliberately containing
  certain community patterns

16
Result of detecting communities
17
Grouping by category

• Group companies in the same category together in
  the layout
• Transaction patterns between category

18
System implementation

• Input
• Pajek .NET format
• Graph visualization package
• PREFUSE
• JUNG

19

• Thanks
• Any questions?