Cooperative Meeting Scheduling among Agents based on Multiple Negotiations

1
Cooperative Meeting Scheduling among Agentsbased
on Multiple Negotiations

• Toramatsu SHINTANI and Takayuki ITO
• Department of Intelligence and Computer Science,
• Nagoya Institute of Technology
• tora_at_ics.nitech.ac.jp
• JAPAN

• Motivation
• Distributed Meeting Scheduler
• Distributed scheduling
• Implementation
• Reaching a Consensus
• Multiple Negotiations (Persuasion Process)
• Preference Revision Using Private Preferences
• Conclusions

2
Motivation
3
Background
4
Distributed Scheduling System
5
The Calendar
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
The Distributed Meeting Scheduling
Deciding attributes of
the meeting and
designing alternatives
10
The Outline of the Persuasion Process
Persuasion between agent A and agent B.
We call A "Persuader" and B
"Compromiser."

1.

A sends a proposal to B.
2.
B tries to revise her preference.
3.
If B could revise her preference, they reach an
agreement.
A
Agent B
A
Agent A
A
A
A
A

f
f
f
f

2
1
3
1
persuade
2
3
11
(No Transcript)
12
Quantifying User's Preference Using Multiple
Attribute Utility Theory
Scheduling a meeting
preference
Size
Convenience
Length
900-1100
1300-1400
900-1000
We can select several options with respect to f
according to the
application area. In our system, we select the
AHP method for
calculating user's utility.

13
Quantifying User's Preference
Using AHP
AHP
The pairwise
comparison
matrix
with respect to the criterion "
Convenience
"
Scheduling a meeting
Weights
900-1000
900-1100
1300-1400
Size
Convenience
Length
1/3
2
0.205
1
900-1000
9
3
1
0.705
900-1100
1/2
1/9
1
0.089
1300-1400
1300-1400
900-1000
900-1100
14
The Preference Revision
15
(No Transcript)
16
The Feature of the Preference Revision
The MC principle
An agent should change an user's preference as
minimal as
possible
The OC principle
An agent should change an user's preference based
on the
preference order of alternatives
In our system, a compromiser tries to adjust
attribute values based on
"
generate and test
"
style. The problem is that the solution space
is too
huge to revise agent's preference.
17
Implementation
18
The Main Features of MiLog
19
(No Transcript)
20
An Example of MiLog
21
Experimental Result
22
Conclusions
A
new multi-agent negotiation

The multiple negotiations can reflect user's
individual preferences.
The preference revision effectively find a
solution for a
compromiser in the persuasion process.

The Distributed Meeting Scheduler

realizing a cooperative meeting scheduling among
agent
improving a trade-off between "reaching a
consensus" and
"reflecting users' preference" in a social
decision.
The result shows that the multi-agent negotiation
based on private
preference is an effective method for a
distributed meeting scheduler.
The process can facilitate reaching a consensus
among agents.