Two new research papers published on arXiv explore the issue of popularity bias amplification in recommender systems. The first paper introduces SPRINT, a method to mitigate this bias in transformer-based recommenders by regularizing attention scores and feed-forward parameters, showing improved accuracy and fairness. The second paper identifies a more general cause of popularity bias in dot-product recommender scores, attributing it to the item-marginal term rather than solely to transformer architectures, and demonstrates significant reduction in popularity-aligned score energy by separating this term. AI
IMPACT These papers offer new techniques to improve fairness and accuracy in AI-powered recommendation systems, potentially leading to better user experiences and a healthier content ecosystem.
RANK_REASON Two academic papers published on arXiv detailing new findings and methods related to recommender systems.
Read on arXiv cs.IR (Information Retrieval) →
- Alibaba
- alphaXiv
- arXiv
- CatalyzeX
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- Matthew effect
- ScienceCast
- SPRINT
- transformer
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
