From Leakage to Fidelity: Reliable Benchmarking for Temporal Cascade Prediction
The episode discusses the paper 'From Leakage to Fidelity,' which improves temporal cascade prediction by addressing data leakage and computational inefficiency. The authors propose using time-ordered splitting, the e-commerce Taoke dataset, and a lightweight model called CasTemp to achieve highly accurate, yet scalable, predictions for real-world conversion outcomes.