Multi-Task Learning

• General Recommender Systems

  • Use Case: Improving personalization by modeling multiple user behaviors.

  • Architecture: Shared-bottom, MMoE, PLE

  • Reference: Kumar, “An Introduction to Multi-Task Learning based Recommender Systems”

• Video Recommendation (e.g., YouTube)

  • Use Case: Predicting multiple user engagement metrics like watch time, likes, and shares.

  • Architecture: MMoE (Multi-gate Mixture-of-Experts)

  • Reference: Zhao et al., “Recommending What Video to Watch Next: A Multitask Ranking System”

• E-commerce

  • Use Case: Joint modeling of click-through rate (CTR), conversion rate (CVR), and gross merchandise value (GMV).

  • Architecture: DTN (Deep Multiple Task-specific Feature Interactions Network)

  • Reference: Bi et al., “DTN: Deep Multiple Task-specific Feature Interactions Network for Multi-Task Recommendation”

• Short Video Platforms

  • Use Case: Enhancing recommendations by modeling dynamic user behaviors.

  • Architecture: User Behavior-Aware Multi-Task Learning Model (UBA-SVR)

  • Reference: “A user behavior-aware multi-task learning model for enhanced short video recommendation”

• Advertising

  • Use Case: Modeling sequential user actions such as impression → click → application → approval → activation.

  • Architecture: AITM (Adaptive Information Transfer Multi-task Learning)

  • Reference: “Modeling the Sequential Dependence among Audience Multi-step Conversions with Multi-task Learning”

• Multi-Domain Recommendation

  • Use Case: Handling recommendations across multiple domains and tasks simultaneously.

  • Architecture: M3oE (Multi-Domain Multi-Task Mixture-of-Experts)

  • Reference: Zhang et al., “M3oE: Multi-Domain Multi-Task Mixture-of-Experts Recommendation Framework”