HERO: Diverse-Teacher On-Policy Distillation for Generative Recommendation
What it is
A research paper proposing HERO, a framework that mitigates popularity bias in generative recommendation systems using on-policy distillation. It employs heterogeneous teachers specialized in popular and unpopular items and routes training samples accordingly to improve diversity.
The problem it solves
Existing on-policy distillation methods in generative recommendation exacerbate popularity bias, leading to reduced recommendation diversity and under-exposure of unpopular items.
What you could use it for
- Improving diversity in e-commerce recommendation engines
- Mitigating popularity bias in generative recommendation models
- Deploying diverse recommendation strategies in industrial platforms like 1688
Why it matters
Could significantly increase recommendation diversity and gross merchandise value (GMV) by ensuring unpopular items are adequately represented in model training and inference.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.