Privacy-Preserving Link Prediction
D Demirag, M Namazi, E Ayday, J Clark
A research group at Concordia University Montreal led by Didem Demirag (2022) described privacy-preserving link prediction. Link prediction is a method used to discover important connections between nodes in a graph. The analysis of these linkages helps the data holder to forecast future connections that might emerge between the nodes and to predict missing links in the data.
Some common applications include: (i) social networks, to recommend links between users; (ii) e-commerce or personalized advertisement, to recommend products to users; (iii) in telecommunications, to build optimal phone usage plans between users; and (iv) in bioinformatics, to predict associations between diseases and attributes of patients or to discover associations between genes and different functions.
It can be used in various applications such as social networks, e-commerce, telecommunication, and bioinformatics. Distributed link prediction allows for building connections between users and products based on the preferences of similar users. A cryptographic solution is proposed to ensure privacy in distributed link prediction between multiple graph databases. The solution involves additively homomorphic encryption and private set intersection (PSI). The proposed scheme allows one graph database to hide the identities of nodes from another graph database.
Demirag, D., Namazi, M., Ayday, E., & Clark, J. (2022, September). Privacy-Preserving Link Prediction. In International Workshop on Data Privacy Management (pp. 35-50). Cham: Springer International Publishing.