AdCSP — Cooperative Ad Auctions with Privacy Guarantees

This paper discusses a cooperative model for advertisement bidding with privacy guarantees. In our application, a set of participants need to agree on an allocation of advertisements to be shown on an Internet stream. The participants have preferences that pose constraints to what should be streamed, and define the value of the ads. Streaming an ad to a specific audience yields larger value than streaming it to a random audience. We model the problem using distributed constraint optimization problems (DCOPs), with privacy. Each user has a set of shared variables, and private constraints on those variables. In this context, a local CSP describes a set of interests and constraints on those interests: all participants have stake in the ads being useful, not simply ignored.