Recommender Systems Under Privacy Protection

The pervasive use of personal data by modern algorithmic recommender systems has sparked public outcry for tighter privacy regulations. Personal preferences over different product offerings are a basic constituent of consumer privacy. We study how a profit-driven online platform, who has proprietary information about its products' profitability, designs its recommender system in response to three modes of privacy protection: (i) unprotected privacy, whereby the platform can fully observe the user's personal preference; (ii) privacy protected by an opt-out option that allows the user to choose whether to share her true preferences with the platform; and (iii) privacy protected by a self-disclosure option that allows the user to disclose her preferences, not necessarily truthfully, to the platform. We show that the opt-out privacy protection does not hinder the platform's profitability nor the user's surplus when compared to unprotected privacy. In fact, the platform's optimal policy under unprotected privacy suffices to generate opt-in incentives for the users. Under this policy, the platform should personalize its recommendations when the more profitable product is sufficiently abundant in the market; otherwise, the platform resorts to a non-personalized policy, which always recommends the product to the user indiscriminately. However, when the user's privacy is protected by the self-disclosure option, the platform may have to distort this policy either upward or downward, and we fully characterize those distortions. In particular, the downward distortion leads to algorithmic discrimination, whereby the disadvantaged minority in the society are restricted or deprived of access to potential valuable opportunities. As a result, the self-disclosure protection may lower the societal surplus, despite offering users the highest autonomy over their privacy among the three protection modes.