Review: Agents in Electronic Commerce: Component Technologies for Automated Negotiation and Coalition Formation

This paper reviews six component technologies for automated negotiation and coalition formation.

1. OCSM-contracts in marginal cost based contracting
 Different agents may have different resource so they have different abilities to handle the tasks. A model which allow agents to reallocate tasks.  Agents make contracting decisions based on marginal cost (I am not very sure what is the marginal cost here). The model also defines how agents reallocate task and how to pay for the changing. The author also gives the proof that the globally optimal task allocation can be reached by any hill-climbing algorithm in a finite number of steps. However, the time cost on a large-scale system is still a problem. [2] gives experimental results.

2. leveled commitment contracts
 The contract is binding among agents in traditional negotiation protocols. However, there are events may cause agents decommit. Self-interest agents will decommit when they can get more utility and they will try to avoid the penalty. Thus, the contracts should be able to avoid this situation. Leveled commitment contracts implement the backtracking method for avoiding local optima.

3. anytime coalition structure generation with worst case guarantees
 This part discuss the superadditive and how to guarantee the worst cast is within a bound.

4. trading off computation cost against optimization quality within each coalition

5. distributing search among insincere agents
A method for distributing search works among self-interest agents without a trusted third party.

6. unenforced contract execution.

For 4 & 6, I think I need to read again before I write something...



[1] T. Sandholm “Agents in Electronic Commerce: Component Technologies for Automated Negotiation and Coalition Formation” Autonomous Agents and Multi-Agent Systems, 3(1), 73-96. 2000.

[2] M. R. Andersson and T. W. Sandholm, ‘‘Time-quality tradeoffs in reallocative negotiation with combinatorial contract types,’’ in Proceedings of the National Conference on Artificial Intelligence (AAAI), Orlando, FL, 1999, pp. 3-10.

Review: A Classification Scheme for Negotiation in Electronic Commerce

In the paper "A Classification Scheme for Negotiation in Electronic Commerce", the authors raised problems of "first-generation" system and also give the definition of "Negotiation Space" in order to clarify the possible parameters which may be considered in the space.

The first generation systems only provide users some choices to select.Even though there may be some shopping assistants which can help users look for the best deal, the actions left for users are still limited.

Only select/accept choices mode is not enough for implementing complicated negotiation. By defining the negotiation space, it promotes a clearly framework for next generation negotiation systems. The negotiation space indicated the possible parameters may used in negotiation procedure. Moreover, the negotiation space allows automatic negotiation mechanisms to reach the best deals. More actions for the agents, such as to generate offers, to accept/decline offers, can be implemented to the systems.

Although the negotiation space concept provides a good classification scheme for negotiation, this paper had not give any practical implementation, but three related papers may provide more information about it. They are:

Paurobally and Cunningham

Vetter and Pitsch

Oliveira and Rocha



Reference

E. Oliveira and A.-P. Rocha. Agents advanced features for negotiation in electronic
commerce and virtual organisation formation process. In C. Sierra and F. Dignum,
editors, This book. Springer Verlag, June 2000.

Lomuscio, A. R., M. Wooldridge, and N. R. Jennings. (2001). “A Classification Scheme for Negotiation in Electronic Commerce,” in F. Dignum and C. Sierra (eds.), Agent-Mediated Electronic Commerce: A European Perspective. Springer Verlag, 19–33.

M. Vetter and S. Pitsch. Towards a flexible trading process over the internet. In C. Sierra and F. Dignum, editors, This book. Springer Verlag, June 2000.

S. Paurobally and J. Cunningham. Formal models for negotiation using dynamic logic. In C. Sierra and F. Dignum, editors, This book. Springer Verlag, June 2000.

Some Understandings about Forming Coalitions

A precondition of forming coalitions is that there is a way to distribute the utility and pay off among the coalition members.

Generally thinking the aim of forming a coalition is to gain more social utility. However, for a self-interested agent, it decided to form or join a coalition if and only if the coalition can maximized its utility. A self-interested agent doesn't care about the amount of social utility.

However, is it that all ways to increase the utilities among a group of agents can be seen as forming coalitions.

For example, there are two agents i and j. i wants a good C and j wants a good D. i has the good D and j has the good C. Both agents can maximize their utilities by changing the goods they own. Do i and j form a coalition? I was thinking they do, but now, I think I probably wrong in the way I am thinking it.

The coalition theory is not to concern about how agents cooperate, but it care about the largest value that the coalition can obtain. 


In other words, the concepts of coalition and cooperate are at different levels.

A Review of Zeuthen Strategy

Risk

At step t, agent i decides not to make a concession. He will take a risk if its opponent does not make a concession as well and they will run into a conflict.

Rosenschein & Zlotkin provided a way to measure the risk:

They also described it as:

Risk(i,t) = (utility agent 1 loses by conceding and accepting agent 2's offer) / 

                 (utility agent 1 loses by not conceding and causing a conflict)

We can see from the risk function that the more concession you make, the more risk you take. 

Zeuthen Strategy


Zeuthen Strategy proposes to balance the risks of agents in negotiation. In each step, the agent with a less risk should make a sufficient concession, until they make an agreement.


A Sufficient Concession is one that change the balance of risk between two agents. The agent which has made the sufficient concession should have higher risk than its opponent.


By using Zeuthen Strategy, agents may make an agreement which can maximal the product of utilities (Nash Product), but not the sum of utilities. (Pareto optimal)




The computationally complexity of Zeuthen Strategy is O(2^n).

Hello Blog!

It's a little bit late to say hello to this blog.
Once created 1 year ago, never update...
Anyway, It start again.