A differential privacy framework for matrix factorization recommender systems

Friedman, Arik; Berkovsky, Shlomo; Kâafar, Mohamed Ali

doi:10.1007/s11257-016-9177-7

Cited by 89 publications

(49 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In a specific discussion of recommender systems that use matrix factorization as an underlying technology, Friedman, Berkovsky, and Kaafar (2016) found that "of all the algorithms considered, input perturbation results in the best recommendation accuracy, while guaranteeing a solid level of privacy protection against attacks that aim to gain knowledge about either specific user ratings or even the existence of these ratings" (p. 425). In a specific discussion of recommender systems that use matrix factorization as an underlying technology, Friedman, Berkovsky, and Kaafar (2016) found that "of all the algorithms considered, input perturbation results in the best recommendation accuracy, while guaranteeing a solid level of privacy protection against attacks that aim to gain knowledge about either specific user ratings or even the existence of these ratings" (p. 425).…”

Section: Related Literature On Privacy and Overviewmentioning

confidence: 99%

Recommendations and privacy in the arXiv system: A simulation experiment using historical data

Menkov

Ginsparg

Kantor

2019

Asso for Info Science & Tech

View full text Add to dashboard Cite

Recommender systems may accelerate knowledge discovery in many fields. However, their users may be competitors guarding their ideas before publication or for other reasons. We describe a simulation experiment to assess user privacy against targeted attacks, modeling recommendations based on co-access data. The analysis uses an unusually long (14 years) set of anonymized historical data on user-item accesses. We introduce the notions of "visibility" and "discoverability." We find, based on historical data, that the majority of the actions of arXiv users would be potentially "visible" under targeted attack. However, "discoverability," which incorporates the difficulty of actually seeing a "visible" effect, is very much lower for nearly all users. We consider the effect of changes to the settings of the recommender algorithm on the visibility and discoverability of user actions and propose mitigation strategies that reduce both measures of risk.

show abstract

Section: Related Literature On Privacy and Overviewmentioning

confidence: 99%

Recommendations and privacy in the arXiv system: A simulation experiment using historical data

Menkov

Ginsparg

Kantor

2019

Asso for Info Science & Tech

View full text Add to dashboard Cite

show abstract

“…However, applying LDP to matrix factorization is not trivial. Some existing works of differentially private matrix factorization [3], [21] assume DC is trusted; they do not fit the LDP model. [24], [42] apply LDP on matrix factorization.…”

Section: Our Solutions In a Nutshellmentioning

confidence: 99%

Truth Inference on Sparse Crowdsourcing Data with Local Differential Privacy

Sun

Dong

Wang

et al. 2018

2018 IEEE International Conference on Big Data (Big Data)

View full text Add to dashboard Cite

Crowdsourcing has arisen as a new problem-solving paradigm for tasks that are difficult for computers but easy for humans. However, since the answers collected from the recruited participants (workers) may contain sensitive information, crowdsourcing raises serious privacy concerns. In this paper, we investigate the problem of protecting answer privacy under local differential privacy (LDP), by which individual workers randomize their answers independently and send the perturbed answers to the task requester. The utility goal is to enable to infer the true answer (i.e., truth) from the perturbed data with high accuracy. One of the challenges of LDP perturbation is the sparsity of worker answers (i.e., each worker only answers a small number of tasks). Simple extension of the existing approaches (e.g., Laplace perturbation and randomized response) may incur large error of truth inference on sparse data. Thus we design an efficient new matrix factorization (MF) algorithm under LDP. We prove that our MF algorithm can provide both LDP guarantee and small error of truth inference, regardless of the sparsity of worker answers. We perform extensive experiments on real-world and synthetic datasets, and demonstrate that the MF algorithm performs better than the existing LDP algorithms on sparse crowdsourcing data.

show abstract

“…For recommendations purposes, the algorithm factorises the matrix R u * i into two latent matrices: (i) the user-factor matrix P ; and (ii) the item-factor matrix Q. Equation 5 represents this factorisation where each row p u of P or q i of Q represents the relation between the corresponding latent factor and the user u or item i, respectively, and λ regularises the learned factors (Friedman et al 2016).…”

Section: Rating Predictionmentioning

confidence: 99%

Profiling And Rating Prediction From Multi-Criteria Crowd-Sourced Hotel Ratings

Leal¹,

González–Vélez²,

Malheiro³

et al. 2017

ECMS 2017 Proceedings Edited by Zita Zoltay Paprika, Péter Horák, Kata Váradi, Péter Tamás Zwierczyk, Ágnes Vidovics-Dancs, Ján

View full text Add to dashboard Cite

Based on historical user information, collaborative filters predict for a given user the classification of unknown items, typically using a single criterion. However, a crowd typically rates tourism resources using multi-criteria, i.e., each user provides multiple ratings per item. In order to apply standard collaborative filtering, it is necessary to have a unique classification per user and item. This unique classification can be based on a single rating -single criterion (SC) profiling -or on the multiple ratings available -multicriteria (MC) profiling. Exploring both SC and MC profiling, this work proposes: (ı) the selection of the most representative crowd-sourced rating; and (ıı) the combination of the different user ratings per item, using the average of the non-null ratings or the personalised weighted average based on the user rating profile. Having employed matrix factorisation to predict unknown ratings, we argue that the personalised combination of multi-criteria item ratings improves the tourist profile and, consequently, the quality of the collaborative predictions. Thus, this paper contributes to a novel approach for guest profiling based on multi-criteria hotel ratings and to the prediction of hotel guest ratings based on the Alternating Least Squares algorithm. Our experiments with crowd-sourced Expedia and TripAdvisor data show that the proposed method improves the accuracy of the hotel rating predictions.

show abstract

A differential privacy framework for matrix factorization recommender systems

Cited by 89 publications

References 38 publications

Recommendations and privacy in the arXiv system: A simulation experiment using historical data

Recommendations and privacy in the arXiv system: A simulation experiment using historical data

Truth Inference on Sparse Crowdsourcing Data with Local Differential Privacy

Profiling And Rating Prediction From Multi-Criteria Crowd-Sourced Hotel Ratings

Contact Info

Product

Resources

About