Evolving multi-objective ranking models for GMV optimization in E-Commerce

In this paper, we focus on the challenge of GMV optimization in an e-commerce marketplace. We develop a production grade ranking system capable of evolving neural networks to efficiently trade off between different components while minimizing the impact on conversion. We further validate our approach with both offline analysis and online A/B experiments in a large scale production search engine. Finally, we contribute our production source code to open source to encourage further exploration in this space.