Sampling-Bias-Corrected Neural Modeling for Large Corpus Item Recommendations
Time:September2019Author:XinyangYi,JiYang,LichanHong,DerekZhiyuanCheng,LukaszHeldt,AditeeKumthekar,ZheZhao,LiWei,EdChiAbstract工业界目前训练双塔结构一般是通过随机mini-batch的方式来优化损失函数(in-batchnegatives)。这种训练方式存在一个显著问题就是