It is meaningful and essential to study and ascertain how to automatically organize user’s friends and discover social circles [3] [4], which could help solve many practical problems. For example, a conference organizer may better target potential participants by issuing invitations around a core group of known experts; a common person wants to read the latest news from his colleagues instead of scrolling through all the news from other users [5]. However, most of current social circle identification methods [3] [4][5][6][7] lacks emphasis on data set quality and they could not predict well when there is missing value (evidence) in the query.