Private and Utility Enhanced Recommendations with Local Differential Privacy and Gaussian Mixture Model

Recommendation systems rely heavily on users behavioural and preferential\ndata (e.g. ratings, likes) to produce accurate recommendations. However, users\nexperience privacy concerns due to unethical data aggregation and analytical\npractices carried out by the Service Providers (SP). Local differential privacy\n(LDP) based perturbation mechanisms add noise to users data at user side before\nsending it to the SP. The SP then uses the perturbed data to perform\nrecommendations. Although LDP protects the privacy of users from SP, it causes\na substantial decline in predictive accuracy. To address this issue, we propose\nan LDP-based Matrix Factorization (MF) with a Gaussian Mixture Model (MoG). The\nLDP perturbation mechanism, Bounded Laplace (BLP), regulates the effect of\nnoise by confining the perturbed ratings to a predetermined domain. We derive a\nsufficient condition of the scale parameter for BLP to satisfy $\\epsilon$ LDP.\nAt the SP, The MoG model estimates the noise added to perturbed ratings and the\nMF algorithm predicts missing ratings. Our proposed LDP based recommendation\nsystem improves the recommendation accuracy without violating LDP principles.\nThe empirical evaluations carried out on three real world datasets, i.e.,\nMovielens, Libimseti and Jester, demonstrate that our method offers a\nsubstantial increase in predictive accuracy under strong privacy guarantee.\n