Group decision making (GDM) problems require consensus reaching processes; however, these can be time consuming and costly. As experts change their evaluations after exchanging opinions and being influenced by others, these influences are spread across the various expert trust relationships. Because of the experts’ knowledge limits, the evaluations on the alternatives and the trust relationships are generally described using probabilistic linguistic terms. Therefore, to simplify the decision making process and avoid decision bias, this paper proposes a particle swarm optimization method that incorporates a trust relationship based social network for GDM under a probabilistic linguistic environment. Each expert is regarded as a particle that moves toward the final evaluation and reaches the threshold. A fitness function is built to measure the consensus levels, and the updated function is improved by the trust relationships to derive the new evaluations. A numerical example is then given to illustrate the feasibility of the proposed approach and comparisons given to further elucidate its novelty and validity.