The ever-growing amount of data available on the Internet calls for personalization. Yet, the most effective personalization schemes, such as those based on collaborative filtering (CF), are notoriously resource greedy. We argue that scalable infrastructures should rely on P2P design to scale to that increasing number of users, data and dynamics.
I will present a novel scalable k-nearest neighbor protocol, which P2P flavor provides scalability by design. This protocol provides each user with an implicit social network composed of users with similar tastes in a given application.

This protocol has been instanciated in various settings:
(1) A P2P system, WhatsUp, a collaborative filtering system for disseminating news items in a large-scale dynamic setting with no central authority;
(2) A hybrid recommendation infrastucture HyRec, an online cost-effective scalable system for CF personalization, offloading CPU-intensive recommendation tasks to front-end client browsers, while retaining storage and orchestration tasks within back-end servers;
(3) A cloud-based centralized recommendation engine.

Experiment show that our solution outperforms alternatives with respect to cost while maintening the quality of personalization.