Certainty versus Actuality in Trust Networks

From Master Projects
Jump to: navigation, search

has title::Discounting Experience in Referral Networks
status: finished
Master: project within::Computational Intelligence and Selforganisation
Student name: student name::Nicolas Höning
Start start date:=2009/04/01
End end date:=2009/08/27
Supervisor: Martijn Schut
Second reader: has second reader::Guszti Eiben
Company: has company::VU
Poster: has poster::Media:Media:Posternaam.pdf

Signature supervisor



KIM 1: Abstract

Presentation date: June 05, 2009

A recent stream of research (Josang, Singh, Wang) developed a certainty-based trust representation, which not only communicates as how likely a good experience with the trustee is estimated, but also on how much actual experience this opinion is based. Agents can then rely more on trust reports which contain more information, thus interpreting them as more certain. In an environment where service providers may change the quality of their service, agents now face a dilemma: They desire two things in a trust report, certainty and actuality, but those are antiproportional: To make a trust report more actual, old information has to be discounted.

This research will build upon work by Hang et al (2008), where a simple trust referral network was designed to test certainty-based trust referral algebras in the presence of dynamic service quality and misleading referrals. Very basic time-based discounting of information has been implemented, but the above mentioned dilemma has not been explored. I plan to extend the network design to examine the effect of discounting strategies of referrers on the network performance.

KIM 2: Abstract

Presentation date: August 27, 2009

In the context of information systems, a disruptive environment demands a solution to a trade-off: How quickly should agents forget experience? If they cherish their memories, they can build their decisions on larger data sets; if they forget quickly, they can respond well to change. This task can be characterised as a decentralised learning problem and its solution highly depends on the environment. In this work, we establish a testbed to examine this problem by building on work by Hang et al (2008). We observe which forgetting patterns work best and what happens if agents choose their forgetting rate freely.