Assessing trust relationships in social networks is vital for a lot of applications. However, the problem is challenging as trust relationships are often not directly observable. The goal of this paper is to predict trust relationships not yet observed based on the observed trust network. The authors follow the work of Guha et al. [1] and propose a probabilistic trust propagation model.

The authors reduced the set of propagation mechanisms in Guha et al. [1] to two fundamental ones in transposition and forwarding, and reformulated the problem as a statistical inference problem: the observed and unobserved trust relationships are modeled as known and unknown variables. The propagation mechanisms are treated as potential functions; the joint probability of observed and unobserved trust relationships is modeled as the product of potential functions. The authors applied belief propagation to solving the proposed optimization problem.

The authors used the benchmark datasets from Ciao and Epinions, and compared the prediction accuracy of their method with that in Guha et al. [1]. The proposed model considerably outperforms previous methods in terms of top-K precision.

To conclude, the new probabilistic prediction model shows superior performance over existing ones. However, this method only considers the propagation of trust relationships (for example, once formed, trust relationships exist forever), while the interactions between trust and distrust relationships are not considered. This may limit its applicability.

Aside from this issue, the paper is very well written and provides extensive theoretical analysis and experimental results.