Social networks are gaining importance as an important application landscape for machine learning techniques. Currently, it is a prevalent thinking to introduce social networks into group decision-making (GMD) research, despite exhibiting promising performance, there remain significant challenges, including 1) trust risk is often overlooked, and 2) group opinions are excessively relied upon. In this regard, this paper introduces the trust risk test for group consensus (SN-TRT-GC) by using probabilistic linguistic preference relations (PLPRs) under probabilistic linguistic term sets (PLTSs). The proposed method ensures that the experts competence matches the discourse as much as possible by virtue of the trust risk value. Moreover, the opinion evolution relies on the dual attributes of similarity and trust, which promotes the exchange of opinions among experts and greatly accelerates the consensus reaching process (CRP). Empirical investigations of satisfaction surveys demonstrate the validity and superiority of the presented SN-TRT-GC technique.