Product Recommendations for New Users
Recommender systems suggest products based upon users' preferences. But what if the user is new, meaning there isn't enough information to make good recommendations? The method from this study quantifies inter-users dependency, and helps new users make decisions by referring to existing users' decision-making process.
Recommender systems have drawn great attention because of their utility in many areas, including movies reviews, restaurant selection, financial services, and even identifying gene therapies. As data from both users and items grows exponentially, there is a great demand to develop efficient recommender systems that track users’ preferences and recommend potential items of interest.
In this study, Xuan Bi and his co-authors propose a group-specific method that accommodates dependency among users and among items to make recommendations. They also propose a new computing solution, such that the method can be implemented on systems involving large-scale customer records. This solution avoids large matrices operation and big-memory storage, and therefore makes it feasible to achieve scalable computing.
The researchers' numerical studies indicate that the proposed method improves prediction accuracy significantly, especially when the vast majority of active users are new users.
This research contributes to an effective recommender system, which accommodates the fact that many users' decisions depend upon others. Broadly speaking, the proposed method helps businesses effectively identify users, especially new users, who may be interested in certain products.
Methods & Tools
- Advanced statistics and machine learning techniques (matrix factorization and group-specific random effects)
- R, Python, Matlab
Read the Paper
in the Journal of the American Statistical Association