- A group of people is often required to reach a joint decision and choose a single activity in which they will all participate. Members of such group often interact via online social networks. Group decision making requires knowledge of members’ preferences; however, in many cases the members’ preferences are not fully available. We consider a scenario where preferences are known only partially and present a group decision support framework that determines a single winning item using a voting procedure while minimizing the number of queries for members’ preferences. The framework uses a probabilistic algorithm based on the social similarity among the group members. To evaluate our framework we have built LetsDoIt, a decision support system prototype for leisure actives. We compared several types of groups and prediction methods and reveal that for groups with high internal social similarity, the algorithm reaches a decision using fewer queries, and thus less communication with the users is required. Moreover, the runtime of the algorithms using social based prediction methods is less than half that of the algorithm using rating based prediction methods. These results suggest that incorporating social similarity data, when available, can be of value.