Improving the use of advisor networks for multi-agent trust modelling

This paper provides an approach for improving the trust modelling of users when a social network of advisors is employed (for example when advisors are recommending the most trustworthy service providers). We present three important improvements to trust modelling, two directly relating to the size of the network (through either the use of a threshold or by setting a maximum network size) and a third (advisor referrals) which focuses on ensuring that the advisors have attained an appropriate level of expertise, for the advice that they provide. Experimental results confirm the value of our methods when choosing parameters in a principled manner, leading to improvements in the accuracy of trust modelling (shown in the context of electronic marketplaces). In all, this research provides insights into how to set the size and composition of social networks, in order to effectively integrate the advice of peers when modelling the trust of users.