In a follow-up study \cite{Chen_2009}, the aggregation algorithm used by the DYK widget (termed ‘SONAR’) was compared with three other algorithms for people recommendation: (1) Content Matching (CM) — based on cosine similarity of the content created by both users: profile entries, status messages, photos’ text, shared lists, job title, location, description, and tags. Word vectors were created by a simple TF-IDF procedure. Latent semantic analysis (LSA) was not shown to produce better results and was not applied since it does not yield intuitive explanations; (2) Content plus Link (CplusL) — combined CM with social links. A social link was defined as a sequence of 3 or 4 users, where for each pair of users in the sequence u1 and u2, either u1 connects to u2, u2 connects to u1, or u1 commented on u2’s content; (3) Friend of Friends (FoF) — based on the number of mutual friends, as done in many of the popular SNSs. The FoF algorithm was able to produce recommendations for only 57.2% of the users (compared to 87.7% for SONAR). Figure \ref{fig:fig10} shows the recommendation widget.