Evaluation was based on a user survey and a controlled field study. Figure \ref{fig:fig11} shows the main survey results. CM and CplusL produced mostly unknown people, while SONAR and FoF produced mostly known individuals. As could be expected, a higher portion of the recommended people who were familiar to the user were rated as good recommendations and resulted in a “connect” action. Yet, the unknown recommended individuals may help discover new potential friends. The overall superiority of algorithms that involve social links over content was clear: only 30.5% of the CM recommendations resulted in a connect action, compared to 40% for CplusL, 47.7% for FoF, and 59.7% for SONAR.