Task: Design a Reputation System

Design the architecture and the interface of a reputation system.

Purpose

A reputation system is useful in large systems where interactions between specific agents are infrequent.

Relationships

Parent Practices	Trust-based Interaction Design
Roles	Primary Performer: Architect	Additional Performers: Developer
Inputs	Mandatory: Trust Model	Optional: [Multi-Agent System Architecture] [Requirements Model]
Outputs	[Multi-Agent System Architecture] Trust Model

Main Description

In systems in which many agents interact with each other and a high fluctuation of interaction partners can be expected, direct trust based on personal experiences is often too limited. Instead, a reputation system that aggregates trust values created by the agents or collects experiences and transforms them with a trust metric can be used to allow agents to gauge previously unknown partners. Such a system thus helps to spread information about other agents in the system and can be a basis for trustworthy interaction even in cases in which individual pairings are unique occurences.

It is recommended to use existing reputation system architectures whenever possible. Such architectures can be broadly distinguished into centralised and decentralised approaches. In the former case, the reputation system has a central manager that gathers and aggregates data and is the reference for all agents in the system. In the latter case, a number of such agents exist or reputation is calculated in a fully decentralised fashion as a form of gossipping.

Steps

Design the architecture of the reputation system

The literature proposes different architectures for reputation systems that can be reused when designing one for the considered system. Usually, a reputation manager is used that either collects experiences from the agents and processes them itself or it gathers trust values directly which are then aggregated. Such a manager can either be centralised or distributed.

Define the interface for gathering the information necessary to calculate a reputation value

Agents can either contribute their experiences or trust values continuously or the reputation system can request them when they are needed. In both cases, interfaces have to be defined that facilitate this exchange of information.

Define the interface and protocol for querying the reputation of an agent

Agents need to have the possibility to query the reputation of potential interaction partners. The reputation system thus has to provide an appropriate interface and protocol to make this data accessible.

More Information

Concepts	Reputation in Multi-Agent Systems
Whitepapers	Anders et al. - Patterns to Measure and Utilize Trust in Multi-Agent Systems

This material is made available under the Creative Commons—Attribution-ShareAlike License v3.0.

Parts of this material is based on the Eclipse Process Framework Practice Library, distributed under the Eclipse Public License V1.0.