A Trust and Reputation Model for Supply Chain Management

Yasaman HaghpanahIJCAI 2011 · DOI

Abstract

In global supply chain management (SCM), agents must navigate complex trust landscapes using both direct experience and third-party reports. However, reported observations are frequently biased or unreliable. This paper introduces the Cognitive Reputation (CoRe) model, which utilizes Bayesian modeling to identify and "decode" reporter biases rather than simply discounting them. By categorizing reporters as honest, optimistic, or pessimistic, CoRe effectively translates subjective feedback into actionable data. Evaluation against the TRAVOS benchmark shows that CoRe achieves a 35% reduction in identification error and a 9% increase in overall agent payoff.

Introduction

Effective decision-making in multi-agent supply chains relies heavily on trust. Agents (suppliers, distributors, and retailers) must assess the reliability of potential partners. While direct observations provide high-quality data, they are slow and costly to accumulate. Consequently, agents often rely on reputation—reported observations from other agents. The primary obstacle is the inherent subjectivity of these reports; agents may be habitually optimistic, pessimistic, or intentionally deceptive. Traditional models like TRAVOS often discard biased reports as noise, losing valuable information in the process.

The Cognitive Reputation (CoRe) Model

The CoRe model addresses these challenges by treating bias as a signal to be decoded. It integrates several key components:

  • Bayesian Model Averaging: CoRe maintains a probabilistic belief about each reporter's "type" (Honest, Optimistic, or Pessimistic). By comparing a reporter's claims with actual observed outcomes over time, the system updates these probabilities.
  • Bias Translation: Instead of ignoring a biased report, CoRe applies a translation layer. For instance, if a reporter is identified as pessimistic, a mediocre review is interpreted as indicating a high-quality service.
  • Utility-Based Decision Making: The model employs a utility function (Benefit - Cost) to estimate the expected reward of a partnership, allowing agents to make economically rational decisions.
  • Multi-Factor Trust: CoRe is designed for the multi-dimensional nature of SCM, accounting for variables such as product quality, delivery speed, and payment reliability.

Experimental Evaluation

The CoRe model was evaluated using a simulated Iterated Prisoner’s Dilemma (IPD) platform. The simulation included a "RepSeeker" agent interacting with various "Reportees" (service providers) and a diverse population of "Reporters" (honest, optimistic, and pessimistic). The performance was benchmarked against TRAVOS, a state-of-the-art probabilistic trust model.

Key Results

The experimental results demonstrate the superiority of the CoRe approach in both accuracy and economic outcome.

Comparison of mean error between CoRe and TRAVOS models over 100 rounds.

As shown in the figure above, CoRe significantly outperformed TRAVOS in identifying agent types. CoRe achieved a mean error of 0.037, compared to 0.057 for TRAVOS—a 35% improvement. This increased accuracy in trust assessment directly translated into better partnership choices and higher profitability.

Comparison of mean payoff between CoRe and TRAVOS models over 100 rounds.

The average payoff per round increased from 1.84 (TRAVOS) to 2.01 (CoRe), representing a 9% improvement in utility. These results were statistically validated using a t-test, confirming the robustness of the CoRe model's performance gains.

Discussion and Future Work

The core insight of this research is that biased information remains valuable if the character of the reporter is understood. By modeling reporter types, CoRe enables "newcomer" agents to build accurate market profiles much faster than through direct experience alone.

Future research will focus on expanding the model to handle incomplete or irrelevant data, context-dependent reporter behaviors, and high-dimensional trust tracking across multiple performance metrics.

Summary updated

BibTeX
@inproceedings{haghpanah2011trust,
  author = {Yasaman Haghpanah},
  title = {{A Trust and Reputation Model for Supply Chain Management}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2011},
  doi = {10.5591/978-1-57735-516-8/ijcai11-479},
}
Figure