Reproduce “Feature-aware Diversified Re-ranking with Disentangled Representations for Relevant Recommendation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zihan Lin, Hui Wang, Jingshu Mao, Wayne Xin Zhao, Cheng Wang, Peng Jiang, Ji-Rong Wen
The Feature Disentanglement Self-Balancing Re-ranking (FDSB) framework enhances relevant recommendation by learning disentangled feature representations and adaptively balancing relevance and diversity. It demonstrated improved user engagement and diversity metrics in both offline evaluations and online A/B tests on the KuaiShou app.
Key insights
- Disentangled representations can effectively extract fine-grained, semantically distinct aspects from noisy item features, which is crucial for achieving feature-aware diversity.
- An adaptive, aspect-specific balancing mechanism for relevance and diversity outperforms static weighting, allowing for more nuanced trade-offs during recommendation.
- Explicitly modeling feature-level diversity is essential in relevant recommendation scenarios where candidate items are inherently highly similar, helping to mitigate information cocoons.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.