A Privacy-Preserving Friend Recommendation Mechanism for Online Social Networks
Fukang Liu, WU Guo-rui, Yang Liu · 2020
Friend recommendation systems is widely applied in the context of online social networks (OSNs). Such systems aim to expand users' social network to increase users' engagement with related OSNs. However, during the cold-start stage, in which situation no sufficient information could be used to provide recommendation to new users, social relationships among existing users might be disclosed. As existing users' social relationship might be used to provide recommendation for new users. In this paper, we solve such privacy problem by applying a privacy-preserving friend recommendation mechanism. The novelty of this mechanism lies in its combination of deep learning and differential privacy method. The balance of privacy preservation and social recommendation is achieved by introducing node2vec to generate users' latent features, then performing the information fusion with Heterogeneous Information Networks (HIN), and finally using deep neural network (DNN) which accords with differential privacy to make privacy-preserving recommendations. The mechanism is experimented on a real dataset Higgs Twitter Dataset. The result shows that our method can achieve certain balance to obtain good effect of recommendation without disclosing users' privacy.