Reproduce “Efficient Swing Computation for Retrieval in Large-Scale Recommender Systems” 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
Runhao Jiang, Renchi Yang
An adaptive approximation framework makes Swing-based item-to-item retrieval practical for high-degree query items by combining compressed set intersections, grouped sampling, subset sampling, and confidence-guided top-K refinement. Across billion-scale interaction graphs, the method delivers up to 100x speedups over exact Swing for approximate scoring and up to three orders of magnitude for top-K retrieval, while maintaining precision above 99.9% in the reported large-scale tests.
Key insights
- The best approximation strategy depends on query degree and neighboring-user degree: grouped sampling is more useful for high-degree cases, while subset sampling offers steadier costs when individual intersections dominate runtime.
- Efficient cardinality computation can matter more than generating full intersections, with the specialized QFilter++ mode providing up to 3.68x speedups over full intersection materialization.
- Small score-estimation errors rarely change the retrieved set when confidence-guided refinement concentrates computation around the K-th ranking boundary.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.