How FIN Boosts CTR in Online Food Ordering: A Spatial‑Temporal Modeling Breakthrough

The paper introduces FIN (Fragment and Integrate Network), a novel spatial‑temporal model that extracts multiple sub‑sequences from ultra‑long user behavior logs, applies simplified and multi‑head attention, and fuses them with physically meaningful set operations, achieving up to 5.7% CTR lift and 7.3% RPM improvement in real‑world food‑delivery advertising.

CTR predictionLong Sequence Modelingai

0 likes · 23 min read

How FIN Boosts CTR in Online Food Ordering: A Spatial‑Temporal Modeling Breakthrough

Network Intelligence Research Center (NIRC)

Aug 22, 2023 · Artificial Intelligence

LONGNET: Extending Transformers to Over 1 Billion Tokens

LONGNET introduces dilated attention to enable Transformers to process sequences exceeding one billion tokens with linear computational cost, preserving performance on shorter inputs and demonstrating strong results on long‑sequence modeling and standard language tasks.

Dilated AttentionLONGNETLanguage Modeling

0 likes · 6 min read

LONGNET: Extending Transformers to Over 1 Billion Tokens

DataFunTalk

Oct 24, 2022 · Artificial Intelligence

Efficient Target Attention (ETA) for Long-Term User Behavior Modeling in Click‑Through Rate Prediction

Efficient Target Attention (ETA) introduces a low‑cost hash‑based attention operator that enables end‑to‑end modeling of ultra‑long user behavior sequences for CTR prediction, achieving significant online CTR, GMV, and QPS improvements in Alibaba’s Taobao feed recommendation system.

Attention MechanismCTR predictionHashing

0 likes · 20 min read

Efficient Target Attention (ETA) for Long-Term User Behavior Modeling in Click‑Through Rate Prediction