Balancing Revenue and Privacy with Signaling Schemes in Online Ad Auctions

In online ad auctions, when an Internet user's certain actions trigger an auction, the auctioneer (the platform) usually sends the information about the user to help the buyers better estimate their valuations. However, by strategically revealing only partial information, we cannot only improve the revenue of the auction, but also help protect the privacy of the user. In this paper, we propose a privacy measure in the online ad auction setting, and seek to maximize a convex combination of revenue and privacy. We formulate the problem as a convex optimization program and derive structural results and properties of the program. We prove that any combination coefficient achieves a certain fraction of the optimal revenue gain and privacy gain, and that we can trade-off between revenue and privacy by simply tuning the combination coefficient. We also show that the gap between the optimal revenue and the revenue achieved by revealing no information can be bounded by a certain valuation discrepancy between the buyers. We also conduct extensive experiments (on both synthetic and real data) to show the effectiveness of our method.