Maria: Multi-Scenario Ranking with Adaptive Feature Learning

This paper proposes Maria, a multi-scenario ranking model with adaptive feature learning for heterogeneous query modalities in e-commerce search. The model addresses the challenge of modeling both commonality and differences across multiple search scenarios (visual search, similar product search, and interest search) through three key modules: Feature Scaling (FS), Feature Refinement (FR), and Feature Correlation Modeling (FCM). Experimental results on both Ali-CCP and Alimama datasets demonstrate superior performance compared to existing methods, effectively mitigating the seesaw effect in multi-scenario learning.