Keynote: Marketplace Modeling: Managing Scale and Accuracy
By Amira Hijazi
Facebook social good works on building powerful tools that improve people’s lives. Fundraiser matching, matching jobs and applicants, blood donation recommendations, friend suggestions, rendering feeds, and online advertising are some of the initiatives announced by Facebook social good. These initiatives have a widespread real-world impact. However, the output of these initiatives needs to be equitable, accurate, and take into consideration the large scale of such initiatives.
In his talk, Dr. Nicolas Stier-Moses from Facebook, discussed some techniques they use at Facebook to enable better decision making. The idea behind these techniques is to create a simple model that can be solved and then use the solution to get insights for the original problem.
Low rank approximation and representative market abstraction are effective methods to find compact representations of marketplace. However, the accuracy of the generated model might be an issue. To overcome this problem, Dr. Moses presented a Bayesian optimization-based model coupled with simulation. To conclude, I believe such methods are very important even outside of the marketplace modeling.