Researchers led by Yong Bai introduced the Efficiently Scalable Transformer (EST) for click‑through rate (CTR) prediction. EST processes all raw inputs in a single sequence, avoiding early aggregation that can lose detail. It adds Lightweight Cross‑Attention to cut redundant interactions and Content Sparse Attention to pick high‑signal behaviors. In tests on Taobao’s display advertising platform, EST beat the existing production model.
Why it matters
Advertisers on Taobao can expect higher earnings per ad impression thanks to the more accurate predictions.
