Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tricolore: Multi-Behavior User Profiling for Enhanced Candidate Generation in Recommender Systems

Xiao Zhou, Zhongxiang Zhao, Hanze Guo

TL;DR

Tricolore tackles the core limitation of single-target optimization in recommender systems by introducing a multi-behavior, multi-vector framework that learns distinct behavior-oriented user embeddings. It combines an elastic multi-bucket encoder, a behavior-wise multi-view fusion module, popularity-balanced negative sampling, and allied multi-task prediction to model global and behavior-specific preferences and their interdependencies. Empirical results on public short-video and e-commerce datasets show state-of-the-art performance, strong cold-start gains, and improved diversity without sacrificing accuracy. The approach offers practical benefits for scalable candidate generation and can be tailored to platform-specific objectives through its flexible multi-task design.

Abstract

Online platforms aggregate extensive user feedback across diverse behaviors, providing a rich source for enhancing user engagement. Traditional recommender systems, however, typically optimize for a single target behavior and represent user preferences with a single vector, limiting their ability to handle multiple important behaviors or optimization objectives. This conventional approach also struggles to capture the full spectrum of user interests, resulting in a narrow item pool during candidate generation. To address these limitations, we present Tricolore, a versatile multi-vector learning framework that uncovers connections between different behavior types for more robust candidate generation. Tricolore's adaptive multi-task structure is also customizable to specific platform needs. To manage the variability in sparsity across behavior types, we incorporate a behavior-wise multi-view fusion module that dynamically enhances learning. Moreover, a popularity-balanced strategy ensures the recommendation list balances accuracy with item popularity, fostering diversity and improving overall performance. Extensive experiments on public datasets demonstrate Tricolore's effectiveness across various recommendation scenarios, from short video platforms to e-commerce. By leveraging a shared base embedding strategy, Tricolore also significantly improves the performance for cold-start users. The source code is publicly available at: https://github.com/abnering/Tricolore.

Tricolore: Multi-Behavior User Profiling for Enhanced Candidate Generation in Recommender Systems

TL;DR

Tricolore tackles the core limitation of single-target optimization in recommender systems by introducing a multi-behavior, multi-vector framework that learns distinct behavior-oriented user embeddings. It combines an elastic multi-bucket encoder, a behavior-wise multi-view fusion module, popularity-balanced negative sampling, and allied multi-task prediction to model global and behavior-specific preferences and their interdependencies. Empirical results on public short-video and e-commerce datasets show state-of-the-art performance, strong cold-start gains, and improved diversity without sacrificing accuracy. The approach offers practical benefits for scalable candidate generation and can be tailored to platform-specific objectives through its flexible multi-task design.

Abstract

Online platforms aggregate extensive user feedback across diverse behaviors, providing a rich source for enhancing user engagement. Traditional recommender systems, however, typically optimize for a single target behavior and represent user preferences with a single vector, limiting their ability to handle multiple important behaviors or optimization objectives. This conventional approach also struggles to capture the full spectrum of user interests, resulting in a narrow item pool during candidate generation. To address these limitations, we present Tricolore, a versatile multi-vector learning framework that uncovers connections between different behavior types for more robust candidate generation. Tricolore's adaptive multi-task structure is also customizable to specific platform needs. To manage the variability in sparsity across behavior types, we incorporate a behavior-wise multi-view fusion module that dynamically enhances learning. Moreover, a popularity-balanced strategy ensures the recommendation list balances accuracy with item popularity, fostering diversity and improving overall performance. Extensive experiments on public datasets demonstrate Tricolore's effectiveness across various recommendation scenarios, from short video platforms to e-commerce. By leveraging a shared base embedding strategy, Tricolore also significantly improves the performance for cold-start users. The source code is publicly available at: https://github.com/abnering/Tricolore.
Paper Structure (29 sections, 15 equations, 5 figures, 8 tables)

This paper contains 29 sections, 15 equations, 5 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Methodology
  5. Multi-behavior Encoder
  6. Behavior-wise Multi-view Fusion
  7. Popularity-balanced Item Representation
  8. Allied Multi-task Prediction
  9. Optimization
  10. Experiments
  11. Experimental Settings
  12. Datasets
  13. Baselines
  14. Evaluation Metrics
  15. Parameters Settings
  16. ...and 14 more sections

Figures (5)

  • Figure 1: An illustration of multiple feedback types in WeChat Channels and the design inspiration of Tricolore.
  • Figure 2: The architecture of the proposed Tricolore framework.
  • Figure 3: Study of custom gate control parameters $f$ on WeChat and Tmall.
  • Figure 4: Study of similarities between individual behavior class embeddings and the base embedding $e_b$.
  • Figure 5: Scoring the Trade-off Between Accuracy and Popularity Metrics.