腾讯QQ看点团队:用迁移学习架构解决短视频冷启推荐问题
导读:新用户和冷用户喜好预测问题一直是推荐系统领域的难题,近期腾讯 QQ 看点(PCG 事业群)团队的一项研究提出了一种迁移学习架构 PeterRec,专门解决新用户和冷用户推荐问题。目前,该论文已被 SIGIR 2020 会议接收。
In fact, it has been recognized that item popularity distribution in most real-world recommendation datasets has a heavy tail, following an approximate power-law or exponential distribution.
Accordingly, most non-positive items drawn by uniform sampling are unpopular due to the long tail distribution, and thus contribute less to the desired loss function. Based on the analysis, it is reasonable to present a popularity-aware sampler to replace the uniform one before performing each SGD.
图 3:Predict 处理
赞 (0)