A Normative Approach to Privacy-Preserving Recommender Systems: Integrating Matrix Factorization and Genetic Algorithms

Table 1

Overview of hyperparameter settings.

Hyperparameters	Role and significance	Application in modeling
Differential privacy parameter (ε)	Controls the level of privacy protection, and smaller ε values indicate stronger privacy protection	Determining the noise size of the differential privacy mechanism
Number of iterations (N)	Controls the number of iterations of the optimization process	Controlling the number of genetic algorithm iterations, i.e., the number of iterations to update the hidden factor matrices U and V
Number of candidate recommendation items (l)	Achieve differential privacy protection through the number of recommendation terms selected by the exponential mechanism	Determine the privacy budget of the selection operation to achieve differential privacy protection of the recommendation results
Mutation step size (η)	Controls the step size of the mutation operation	Control the step size of the genetic algorithm to perturb the hidden factors during the search process
Mutation attenuation factor (β)	Controlling the degree of attenuation of the mutation operation	Controlling the diminishment of the mutation step size during the iteration of the genetic algorithm
Number of genetic algorithm iterations (A)	Controlling the number of iterations of the genetic algorithm	Controlling the number of iterations of the genetic algorithm
Privacy budget for selection operation(ε/2NA)	Realize differential privacy protection for the selection operation	Realize differential privacy protection of the selection operation to protect the privacy of the selection operation of the genetic algorithm