Abstract
In this work, the team formation process in an online game was formulated as a sign prediction problem of social networks. Depending on the participation in either collaborative or competitive teams and squads and four different interaction types among garners, the objective was to predict the signs (positive/negative) of their interactions in colluding or competing teams. Temporal features pertaining to the factors affecting team formation were used in a supervised classification setup. High performance by a classifier denotes that historical information related to garners' prior interactions are effective in predicting the sign of their interactions.