Deep Task-Specific Bottom Representation Network for Multi-Task Recommendation

By Qi Liu et al

Published on Oct. 21, 2023

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

Table of Contents

1. Introduction
2. Related Work
3. Method
3.1 Overview
3.2 Embedding Layer
3.3 TIM: Task-Specific Interest Module
3.3.1 Hypernetwork
3.3.2 Conditional Transformer
4. Conclusion
5. References

Summary

The document discusses the proposal of a Deep Task-specific Bottom Representation Network (DTRN) for Multi-Task Recommendation. It focuses on the challenges faced by Multi-Task Learning (MTL) in Recommendation Systems and introduces the methodology to alleviate negative transfer effects. The task-specific interest module (TIM) and task-specific representation refinement module (TRM) are key components of DTRN. TIM utilizes a Hypernetwork and Conditional Transformer to extract task-specific interests from multiple behavior sequences, while TRM refines the representation for each task. The paper highlights the effectiveness of DTRN through experiments on public and industrial datasets.