Risk interaction changes the probability and impact of a given risk, which may result in a less effective risk response decision (RD). This study presents an approach for supporting the project manager in making RDs, comprising a simulation model of risk interaction network (RIN) for evaluating the RDs and an improved simulated annealing (SA) algorithm for optimizing the RDs. The simulation model considers different risk levels and the corresponding risk interaction cases, which is closer to the reality. In addition to tailoring the SA algorithm to optimize RDs, it is improved through enhancing its neighborhood search with the aid of social network analysis. Specifically, two new network indices are designed for calculating the quantitative significance of RIN elements, i.e. the nodes that denote risks and edges that reflect the risk interactions. The element with a higher significance is more likely to be dealt with when generating a new RD in the neighborhood search. An application is provided to illustrate the utility of the proposed approach; a contrastive analysis of the improved SA and standard SA is also conducted to validate the effectiveness and efficiency of the former.