This paper presents a new approach to optimizing cascade ranking systems, which are widely used in online advertising and recommendation systems. The authors propose the Adaptive Neural Ranking Framework (ARF), which uses multi-task learning to adaptively combine the optimization of relaxed and full targets. This approach is designed to improve the adaptability of optimization targets to data complexities and model capabilities. The authors demonstrate the effectiveness of their method through experiments on a total of 4 public and industrial benchmarks, showing that it has significant application value.
Publication date: 17 Oct 2023
Project Page: https://doi.org/10.1145/1122445.1122456
Paper: https://arxiv.org/pdf/2310.10462
