01/ 01
- Feature engineering: numeric, categorical, sequence, cross features - Model: GBDT (LightGBM / XGBoost) or deep learning (DIN / DIEN / DSSM) - Multi-task learning: predict click + purchase + dwell jointly (MMoE / PLE) - Position bias correction (PAL or inverse propensity weighting) - Calibration and thresholding for display - Ranking service: batch inference with TF-Serving / ONNX Runtime, 100ms budget
自动翻译
自动翻译中…