Fuzzy cognitive maps for dynamic grid service negotiation

Chunyan Miao, Jianshu Weng, Angela Goh, Zhiqi Shen, Bo AnMultiagent Grid Syst 2006 · DOI

TL;DR

This paper introduces a novel approach to dynamic, multi-issue service negotiation in Pervasive Grids by leveraging Fuzzy Cognitive Maps (FCMs). By modeling negotiation issues as interconnected concepts within an FCM, the authors enable agents to account for causal relationships between issues, leading to more agile and lightweight decision-making for faster agreement.

What Problem Does This Paper Address?

The paper addresses the challenges of service trading in the evolving landscape of Grids, specifically the "Economic Grid" and "Pervasive Grid." In the Economic Grid, service providers and consumers negotiate complex agreements on multiple issues like performance and cost. The Pervasive Grid context introduces further constraints: limited computational capacity on devices and unreliable wireless communication, making rapid, automated, and lightweight negotiation crucial. Existing negotiation models often treat negotiation issues independently, failing to capture their inherent interdependencies, and can be computationally intensive, making them unsuitable for pervasive environments.

Methods

The core contribution is the development of a decision-making model for negotiation agents based on Fuzzy Cognitive Maps (FCMs).

  • Fuzzy Cognitive Maps (FCMs): FCMs are used to model the dynamic system of negotiation. Concepts in the FCM represent negotiation issues (e.g., prediction level, response time, cost) and an agent's overall satisfaction. Weights on the arcs between concepts represent the causal relationships and influence between these issues.
  • Dynamic System Modeling: The negotiation process is framed as a dynamic system where agents' mental states (preferences, evaluations) evolve. FCMs are used to infer stable states, which represent a counter-offer that an agent is ready to propose.
  • Lightweight Inference: The FCM inference process is designed to be computationally efficient, allowing agents to quickly reach a stable state and derive a counter-offer.
  • Flexible Acceptance Criterion: Instead of a fixed acceptance threshold, the model uses a flexible criterion: an agent accepts an offer if its evaluation of the received offer is not less than the evaluation of its own potential counter-offer.
  • Weight Assignment: Weights are assigned using a case-based reasoning approach, retrieving similar negotiation contexts from a case base. In the absence of similar cases, default weights are assigned based on linguistic terms representing preferences.

Datasets

Not discussed in the paper. The paper uses a simulated illustrative example of a business forecasting service.

Key Insights and Ideas

  • Interdependencies Matter: The most significant insight is that negotiation issues are rarely independent. Capturing these causal relationships using FCMs leads to more realistic and effective negotiation strategies.
  • FCMs as a Lightweight Negotiation Tool: FCMs offer a computationally efficient way to model complex, multi-issue negotiations, making them suitable for resource-constrained pervasive devices.
  • Mental State as a Dynamic System: Viewing an agent's mental state (preferences, evaluations) as a dynamic system that evolves through FCM inference provides a powerful framework for decision-making.
  • Flexible Acceptance is Key: A dynamic and flexible acceptance criterion, rather than a static one, is crucial for agents to adapt to changing environments and potentially accept offers they might have previously rejected.

Key Results

The paper demonstrates the effectiveness of the FCM-based model through an experimental example.

  • Agreement Reached: The model successfully facilitated a negotiation between a service provider and consumer, reaching an agreement within 10 offer exchanges.
  • Lightweight Property: The average inference iterations to derive a counter-offer were shown to be relatively small (e.g., around 20 iterations for a specific parameter setting).
  • Trade-off in Parameter Selection: The parameter $\lambda$ in the threshold function influences both the speed of inference and the number of offer exchanges. A value of $\lambda=2$ was found to provide a good balance between these two requirements.

Limitations and Future Work

  • Weight Assignment Overhead: The current weight assignment method using case-based reasoning introduces an overhead for maintaining the case base.
  • Formal Analysis: The paper suggests future work to develop a formal way to model weights for easier formal analysis (e.g., Game theoretic analysis) of the model's efficiency and stability.
  • Generalizability: While the model is shown to be promising, further investigation into its applicability across a wider range of negotiation scenarios and environments is needed.

Summary updated

BibTeX
@article{miao2006fuzzy,
  author = {Chunyan Miao and Jianshu Weng and Angela Goh and Zhiqi Shen and Bo An},
  title = {{Fuzzy cognitive maps for dynamic grid service negotiation}},
  journal = {Multiagent Grid Syst.},
  year = {2006},
  doi = {10.3233/mgs-2006-2202},
}
Figure