A framework for privacy-conducive recommendations
Recommendations and advertisements based on consumer behavior patterns are increasingly prevalent, yet carry significant privacy concerns. We propose an easily implemented alternative framework in which publicly available Web data is mined to discover product preference associations. Our framework can be used in a recommender system in lieu of the large customer transaction database that many retailers do not have. Because our approach mines public data, and does so in a coarse manner, it greatly reduces the privacy challenges of behavioral approaches. In addition, because we mine the Web, a dynamic corpus that represents much of human knowledge and sentiment, our algorithms have the potential to automatically adjust for changing preferences. We provide a proof-of-concept evaluation of our framework by testing an instantiation against the Netflix movie dataset and Amazon.com recommendations, and achieve precision and recall far higher than with baselines of randomly selected products or popular products.
Chow, R.; Staddon, J. A framework for privacy-conducive recommendations. Workshop on Privacy in the Electronic Society (WPES); 2010 October 4; Chicago, IL.