Fig. 10

Dependence of SCC size on the number of recommendation candidates. To select personalized recommendations, we first compute the \(N + S\) top non-personalized recommendations for each item. The recommendations are then selected as the ones that have the highest predicted rating for the user. S is a parameter we can choose—the higher S, the more responsive the recommendations become to a user’s preferences. However, increasing S leads to decreasing SCC size. This is caused by an increasing focus on preferred items for the user, leading to one-way structures in the networks. The same effect did not occur for mixed recommendations