Fig. 9.

Bow tie membership over N. Using all personalized recommendations (pure) leads to lower discoverability than using an approach combining personalized and non-personalized recommendations. For personalization, the top $N+75$ recommendation candidates were computed for every item as candidates