Evaluating trustworthiness from past performances: interval-based approaches

Abstract

In many multi-agent systems, especially in the field of e-commerce, the users have to decide whether they sufficiently trust an agent to achieve a certain goal. To help users to make such decisions, an increasing number of trust systems have been developed. By trust system, we mean a system that gathers information about an agent and evaluates its trustworthiness on the basis of this information. The aim of the present paper is to develop, and analyze from an axiomatic point of view, new trust systems based on intervals. More precisely, we assume that a set of grades describing the past performances of an agent is given. Then, the goal is to construct an interval that summarizes these grades. In our opinion, such an interval gives a good account of the trustworthiness of the agent. In addition, this kind of representation format overcomes certain limitations (at a certain cost) of the approaches that represent trustworthiness by a single number. We establish seven axioms that should be satisfied by a summarizing method. Next, we develop two new methods. The first one is based on the idea that certain concentrations of grades are strong enough to pull the bounds of the summarizing interval towards themselves. The second one represents data in the setting of possibility theory, and then computes lower and upper expected values. Finally, we check that our methods satisfy the axioms introduced before, which provide theoretical justifications for them.

A Belief-Based Approach to Measuring Message Acceptability

“…They propose a probabilistic method to represent and define trust as depending on the information gain caused by a piece of evidence-the more information an agent has about an event, the smaller its (probabilistic) uncertainty about that event. Probability theory is also used by Teacy et al [28] to represent trust by taking past interactions with other agents into account while possibility theory is used in [2] for proposing an interval-based representations of trust and distrust based on past performances by considering the fact that data are not necessarily numerous in practice.…”

“…An interpretation I ∈ Ω is a function I : Prop → {0, 1} assigning a truth value p I to every atomic proposition p ∈ Prop and, by extension, a truth value φ I to all formulas φ ∈ L. 2 We will denote by [φ] the set of all models of φ, [φ] = {I : I |= φ}.…”

Weighted logics for artificial intelligence – an introductory discussion

“…Many proposals exist -modal or numerical -where gradedness has been introduced in various ways [85,88,31,112,12]. Still, there is not yet a fully clear view of how graded trust may relate to beliefs and uncertainty.…”