深度学习推荐系统
召回阶段
目标:主要根据用户部分特征,从海量的物品库里,快速找回一小部分用户潜在感兴趣的物品,然后交给排序环节,
Embedding模型之Word2vec
论文地址:word2vec Parameter Learning Explained
Embedding模型之DeepWalk
论文地址:DeepWalk: Online Learning of Social Representations
Embedding模型之LINE
论文地址:LINE: Large-scale Information Network Embedding
Embedding模型之Node2vec
论文地址:node2vec: Scalable Feature Learning for Networks
Embedding模型之EGES
论文地址:Billion-scale Commodity Embedding for E-commerce Recommendation in Alibaba
Embedding模型之YoutubeDNN
论文地址:Deep Neural Networks for YouTube Recommendations
粗排阶段
粗排模型之DNN 双塔
论文地址:Sampling-bias-corrected neural modeling for large corpus item recommendations
精排阶段-单目标
目标:精排环节可以融入较多特征,使用复杂模型,来精准地做个性化推荐。
精排模型之CF
论文地址:Collaborative Filtering Recommender Systems
精排模型之MF
论文地址:Matrix Factorization Techniques For Recommender Systems
精排模型之LR
论文地址:Predicting Clicks: Estimating the Click-Through Rate for New Ads
精排模型之FM
精排模型之GBDT+LR
论文地址:Practical Lessons from Predicting Clicks on Ads at Facebook
精排模型之AutoRec
论文地址:AutoRec: Autoencoders Meet Collaborative Filtering
精排模型之Deep&Cross
论文地址:Deep & Cross Network for Ad Click Predictions
精排模型之NeuralCF
论文地址:Neural Collaborative Filtering
精排模型之PNN
论文地址:Product-based Neural Networks for User Response Prediction
精排模型之Wide&Deep
论文地址:Wide & Deep Learning for Recommender Systems
精排模型之DeepFM
论文地址:DeepFM: A Factorization-Machine based Neural Network for CTR Prediction
官方代码:DeepFM
精排模型之XDeepFM
论文地址:xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems
官方代码:xDeepFM
精排模型之DIN
论文地址:Deep Interest Network for Click-Through Rate Prediction
精排模型之DIEN
论文地址:Deep Interest Evolution Network for Click-Through Rate Prediction
精排阶段-多目标
目标:多目标排序通常是指有两个或两个以上的目标函数,目的是寻求一种排序使得所有的目标函数都达到最优或满意。 在工业界推荐系统中,大多是基于隐式反馈来进行推荐的,用户对推荐结果的满意度通常依赖很多指标(比如,淘宝基于点击,浏览深度(停留时间),加购,收藏,购买,重复购买,好评等行为的相关指标),在不同的推荐系统、不同时期、不同的产品形态下,这些指标的重要程度或者所代表的意义也会有所不同,如何优化最终推荐列表的顺序来使得众多指标在不同的场景下近可能达到最优或者满意,这就是一个多目标排序问题。
多目标模型之SharedBottom
论文地址:An Overview of Multi-Task Learning in Deep Neural Networks
多目标模型之ESMM
论文地址:Entire Space Multi-Task Model: An Effective Approach for Estimating Post-Click Conversion Rate
多目标模型之MMoE
论文地址:Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts
多目标之模型之Youtube
论文地址:Recommending what video to watch next: a multitask ranking system
多目标模型之PLE
参考资料
重排阶段
待更新......