Manipulation-resistant recommender systems

This project developed general techniques for the design of manipulation-resistant recommender systems, as well as specific solutions for applications in which such a recommender could have a significant impact.

One class of applications that this project studied involved Internet sites that utilized user-provided ratings or tags to recommend books, images, sites, or other products to users. Another application researchers focused on was the design of rating aggregation systems for selecting proposals, papers, or job candidates.

Start date: 8/1/2008
End date: 8/31/2012

Read More

Online recommender systems are already widely-deployed as tools to guide users toward items they will like, and the space of potential applications of this technology is even larger. However, there is a growing concern that recommender systems may be manipulated by people with a vested interest in having certain items recommended (or not recommended).  This is exacerbated by the fact that it is often easy for a manipulator to create multiple online accounts in order to carry out an attack. The aim of this project was to research and develop new manipulation-resistant recommender systems that worked within realistic practical constraints.

The project involved formal modeling and theoretical analysis to characterize mechanisms that are provably resistant to manipulation. Simulations were used to study the behavior of the algorithms designed against known attack strategies, and use these to derive new insights and better algorithms. Empirical tests were also conducted on real recommendation datasets to evaluate the performance of the new algorithms the researchers design. The empirical tests measured the actual accuracy loss that resulted from manipulation-resistance mechanisms, and the trade-off between the two.

The research efforts on this study ultimately led to the development of SimRecommender, a modular perl framework to simulate and test attacks on recommender systems. This is an application that could impact many communities positively; its viability depends critically on keeping the aggregate recommendations resistant to manipulation attempts. 

This project received an additional $16,000 through a National Science Foundation REU Supplement to support the work of two undergraduate students.

To get more information about this project, please listen to Rahul Sami discuss it in a YouTube video here.

Grants

Manipulation Resistant Recommender Systems, National Science Foundation: $449,013

REU Supplement: Manipulation Resistant Recommender Systems, National Science Foundation: $16,000


 

The National Science Foundation (NSF) is an independent federal agency created by Congress in 1950 "to promote the progress of science; to advance the national health, prosperity, and welfare; to secure the national defense…"