Privacy risks in recommender systems - Internet Computing, IEEE

R ecommender systems have become important tools in ecommerce. They combine one user’s ratings of products or services with ratings from other users to answer queries such as “Would I like X?” with predictions and suggestions. Users thus receive anonymous recommendations from people with similar tastes. While this process seems innocuous, it aggregates user preferences in ways analogous to statistical database queries, which can be exploited to identify information about a particular user. This is especially true for users with eclectic tastes who rate products across different types or domains in the systems. These straddlers highlight the conflict between personalization and privacy in recommender systems. While straddlers enable serendipitous recommendations, information about their existence could be used in conjunction with other data sources to uncover identities and reveal personal details. We use a graph-theoretic model to study the benefit from and risk to straddlers.