Multi-Scenario Ranking with Adaptive Feature Learning

作     者：Tian, Yu Li, Xubin Zheng, Bo Li, Bofang Deng, Hongbo Wang, Qian Chen, Si Xu, Jian Li, Chenliang 

作者机构：Key Laboratory of Aerospace Information Security and Trusted Computing Ministry of Education School of Cyber Science and Engineering Wuhan University China Alibaba Group China 

出 版 物：《arXiv》 (arXiv)

年 卷 期：2023年

核心收录：

主　　题：Semantics 

摘      要：Recently, Multi-Scenario Learning (MSL) is widely used in recommendation and retrieval systems in the industry because it facilitates transfer learning from different scenarios, mitigating data sparsity and reducing maintenance cost. These efforts produce different MSL paradigms by searching more optimal network structure, such as Auxiliary Network, Expert Network, and Multi-Tower Network. It is intuitive that different scenarios could hold their specific characteristics, activating the user’s intents quite differently. In other words, different kinds of auxiliary features would bear varying importance under different scenarios. With more discriminative feature representations refined in a scenario-aware manner, better ranking performance could be easily obtained without expensive search for the optimal network structure. Unfortunately, this simple idea is mainly overlooked but much desired in real-world systems. To this end, in this paper, we propose a multi-scenario ranking framework with adaptive feature learning (named Maria). Specifically, Maria is devised to inject the scenario semantics in the bottom part of the network to derive more discriminative feature representations. There are three components designed in Maria for this purpose: feature scaling, feature refinement, and feature correlation modeling. The purpose of feature scaling is to highlight the scenario-relevant fields and also suppress the irrelevant ones. Then, the feature refinement utilizes an automatic refiner selection subnetwork for each feature field, such that the high-level semantics with respect to the scenario can be extracted with the optimal expert. Afterwards, we further explicitly derive the feature correlations across fields as complementary signals. The resultant representations are then fed into a simple MoE structure with an additional scenario-shared tower for final prediction. Experiments on two large-scale real-world datasets demonstrate the superiority of Maria against sever