Evaluation was primarily based on a user survey with 290 participants. Figure XXX shows the portion of items rated “interesting” for each of the three network types: familiarity, similarity, and overall. Recommendations from familiar people were found significantly more accurate than recommendations from similar people. The overall network did not improve accuracy on top of the familiarity network. That said, recommendations from similar people were found more diverse and less expected, indicating that the similarity network contributes on other dimensions than accuracy to the recommendation quality.