Hybrid Contrastive Constraints for Multi-Scenario Ad Ranking

By Shanlei Mu et al

Published on July 10, 2017

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Preliminaries
3. Methodology
3.1 Backbone of HC2
3.2 Generalized Contrastive Loss for Capturing Common Knowledge
3.3 Individual Contrastive Loss for Capturing Scenario-Specific Knowledge

Summary

Multi-scenario ad ranking aims at leveraging the data from multiple domains or channels for training a unified ranking model to improve the performance at each individual scenario. The paper proposes a Hybrid Contrastive Constrained approach (HC2) for multi-scenario ad ranking. It introduces a hybrid of generalized and individual contrastive losses to capture commonalities and differences among multiple scenarios. The approach enhances contrastive learning by extending contrastive samples and reweighting them. By incorporating label-aware contrastive samples and diffusion noise enhanced contrastive samples, the model can better capture common knowledge. Additionally, the paper utilizes dropout-based augmentation and cross-scenario encoding for generating meaningful positive and negative contrastive samples to capture scenario-specific knowledge. The proposed HC2 has shown effectiveness in multi-scenario ad ranking through extensive experiments.