Assessment of Trust in Opportunistic Reporting using Belief Functions

This paper addresses the assessment of trust in items reported by opportunistic sources and develops a model taking into account the reliability of sources and the credibility of information. Opportunistic sources are witnesses describing events, scenes or actions of interest and often such sources may also indicate confidence, doubt, skepticism or conviction about the items they report. Reliability captures how trustworthy the analyst believes someone is, by considering that opportunistic sources are not always competent and honest but also erroneous or even malicious. Credibility relies upon self-confidence, a measure of how strongly the source believes what he is uttering. Reliability is furthered decomposed according to competences and skills of the source, intentions and reputation while self-confidence varies in a specified interval. Belief functions are used to formalize the model and trust assessment is implemented thanks to discounting operators. The paper also discusses several experiments designed to investigate the effects of reliability and self-confidence on the quality of reported items, as perceived by the analyst.

I. Introduction

In practical applications, integration of human reporting has much to offer for tasks where subtle interactions and associations are difficult to detect with physical sensors, such as intelligence analysis. Those applications combine semantic information with numerical data and often use opportunistic sources, as different persons observe and then report on actions or events of interest and the information is instantly available.