Recommender systems that make personalized recommendations of items or services to users have become increasingly common e-business tools. Two well known examples are recommendations from Amazon.com and Netflix, which are based on past purchase and ratings, respectively. This WS will (a) explore potential new uses of recommender systems for decision makers and (b) identify algorithmic challenges that need solution to enable these new applications. The science of recommender systems has progressed rapidly since the publication by Resnick and Varian. Such systems find patterns in massive matrices indexed by tens or hundreds of thousands of items and, perhaps, millions of users, but usually with very sparse observed data. Methods typically characterize both items and users using either nearest neighbor methods or matrix factorization. The recent Netflix Prize stimulated important advances in both methods. Another promising method, referred to as matrix completion, has grown out of the new field of compressed sensing.
This WS will explore use of recommender systems in a variety of applications in which they have yet seen much use, including those arising in transportation, homeland security, information distribution and personalized medical treatment. Novel applications of recommender systems may require different algorithms than those developed in the relatively pristine setting of Internet commerce. More complex scenarios may need to efficiently integrate information from multiple modes such as purchase history with other background information about users, requiring a mix of ADT tools, e.g., machine learning, data mining, and search algorithms. An important goal of the WS is to identify a set of requirements for the next wave of recommender systems and to inspire research to fill such needs. Among the topics we will consider are: how to efficiently acquire user profiles (demographic data, preferences, social aspects); evaluation of satisfaction by users; retrieval of personalized information while protecting privacy and security; understanding and trusting recommendations.
Next: Call for Participation
Workshop Index
DIMACS Homepage