1.运用场景
CTR模型/圈定人群推荐/long-tail contents recommend。
2.创新点
design a novel network structure named attention-merge layer.
look-alike learning models global representation and local representation of seeds on global and local attention units.
applies seeds clustering which is processed asynchronously online to reduce the scale of seeds and the time complexity of online prediction on attention units.
3.算法原理
3.1 网络框架
3.2 Real-time Attention Based Look-alike Model for Recommender System
4.算法理解
look-alike模式及用户聚类。