RSpaper-read

RSpaper-read分享一些我读过的经典推荐论文，质量有保障，适合初学者入门。主要分为三部分内容：

1. Review：任何领域入门都少不了综述，推荐的文章包括了基于深度学习的推荐、图学习等；
2. Model：算法模型肯定是推荐领域的重点，按照不同阶段，再细化分为召回（matching）与排序（ranking）：
   • matching：召回阶段的模型面临的数据样本是整个物料库，所以它需要在低延时的前提下完成候选物品集的召回给排序阶段；
   • ranking：排序阶段区别于召回，要求模型更加复杂，重特征之间的交叉，主要的指标是CTR；
3. Others：推荐中其他的方向或者有趣的内容；

Review

以下五篇综述都非常适合入门推荐系统：

	Paper	Published in	Time
[1]	Deep Learning for Matching in Search and Recommendation	SIGIR	2018
[2]	Deep Learning Based Recommender System: A Survey and New Perspectives	ACM Computing Surveys	2019
[3]	Learning and Reasoning on Graph for Recommendation	CIKM	2019
[4]	Graph Learning Approaches to Recommender Systems: A Review	IJCAI	2021
[5]	Sequential Recommender Systems: Challenges, Progress and Prospects	AAAI	2019

 

Model

模型按照按照工业界来划分，召回和排序两个大块，由于粗排的文章没有读过，就先不加在里面了。

matching

召回阶段，工业界一般会采用多路召回的形式，即使是现在经常使用的基于向量化的召回，也只会作为其中的一路。多路召回的模型中，最常用的就是ItemCF（基于实际场景），现在工业界也经常会将其作为一路，毕竟又简单又好用。再往后，最经典的就是Matrix Factorization（矩阵分解），召回、排序都可以应用。

	Paper	Published in	Time
[1]	Factorization Meets the Neighborhood: a Multifaceted Collaborative Filtering Model | SVD++	KDD	2008
[2]	Matrix Factorization Techniques for Recommender Systems|MF	IEEE	2009
[3]	Neural network-based Collaborative Filtering | NCF	WWW	2017

 

再往后，就是基于向量化的召回模型（MF其实也算），双塔模型是其中最为通用的架构之一，下面三篇是具有浓厚工业风的文章，业界应用也非常多。

	Paper	Published in	Time
[4]	Learning Deep Structured Semantic Models for Web Search using Clickthrough Data｜DSSM	CIKM	2013
[5]	Deep Neural Networks for YouTube Recommendations |YoutubeDNN	RecSys	2016
[6]	Sampling-Bias-Corrected Neural Modeling for Large Corpus Item Recommendations	RecSys	2019

 

目前，也有很多通过用户行为序列来表征用户，即序列推荐，这也是我个人的研究方向，文章包含了学术与工业。

	Paper	Published in	Time
[7]	Factorizing personalized markov chains for next-basket recommendation | FMPC	KDD	2010
[8]	Learning hierarchical representation model for nextbasket recommendation|HRM	IEEE	2015
[9]	Translation-based recommendation: A scalable method for modeling sequential behavior | TransRec	IJCAI	2018
[10]	Session-based Recommendation with Recurrent Neural Networks | GRU4Rec	ICLR	2016
[11]	Recurrent neural networks with top-k gains for session-based recommendations | GRU4Rec+	WWW	2017
[12]	Personalized top-n sequential recommendation via convolutional sequence embedding | Caser	ICDM	2018
[13]	Self-Attentive Sequential Recommendation | SASRec	ICDM	2018
[14]	STAMP: short-term attention/memory priority model for session-based recommendation｜STAMP	KDD	2018
[15]	Next item recommendation with self-attentive metric learning｜AttRec	AAAI	2019
[16]	BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer｜BERT4Rec	CIKM	2019
[17]	Multi-Interest Network with Dynamic Routing for Recommendation at Tmall | MIND	CIKM	2019
[18]	FISSA: fusing item similarity models with self-attention networks for sequential recommendation｜FISSA	RecSys	2020
[19]	SSE-PT: Sequential recommendation via personalized transforme｜SSE-PT	KDD	2020
[20]	Time Interval Aware Self-Attention for Sequential Recommendation｜TiSASRec	WSDM	2020
[21]	MEANTIME: Mixture of Attention Mechanisms with Multi-temporal Embeddings for Sequential Recommendation｜MEANTIME	RecSys	2020
[22]	Controllable Multi-Interest Framework for Recommendation | ComiRec	KDD	2020
[23]	S3 -Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization｜S3	CIKM	2020
[24]	User BERT: self-supervised user representation learning｜u-bert	ICLR	2021
[25]	Session-Based Recommendation with Graph Neural Networks｜SR-GNN	AAAI	2019
[26]	Sparse-Interest Network for Sequential Recommendation|SINE	WSDM	2021
[27]	SDM: Sequential Deep Matching Model for Online Large-scale Recommender System|SDM	CIKM	2019

 

ranking

这里的ranking主要指的是精排部分的模型，

	Paper	Published in	Time
[1]	Factorization Machines | FM	ICDM	2010
[2]	Field-aware Factorization Machines for CTR Prediction｜FFM	RecSys	2016
[3]	Wide & Deep Learning for Recommender Systems｜WDL	DLRS	2016
[4]	Deep Crossing: Web-Scale Modeling without Manually Crafted Combinatorial Features | Deep Crossing	KDD	2016
[5]	Product-based Neural Networks for User Response Prediction | PNN	ICDM	2016
[6]	Deep & Cross Network for Ad Click Predictions | DCN	ADKDD	2017
[7]	Neural Factorization Machines for Sparse Predictive Analytics | NFM	SIGIR	2018
[8]	Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks | AFM	IJCAI	2017
[9]	DeepFM: A Factorization-Machine based Neural Network for CTR Prediction | DeepFM	IJCAI	2017
[10]	xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems | xDeepFM	KDD	2018
[11]	Deep Interest Network for Click-Through Rate Prediction | DIN	KDD	2018
[12]	Behavior Sequence Transformer for E-commerce Recommendation in Alibaba | BST	DLP-KDD	2019
[13]	Deep Interest Evolution Network for Click-Through Rate Prediction ｜ DIEN	AAAI	2019
[14]	Deep Match to Rank Model for Personalized Click-Through Rate Prediction | DMR	AAAI	2020

序列推荐：

	Paper	Published in	Time
[15]	Sequential recommendation with user memory networks｜MANN	WSDM	2018

多任务：

	Paper	Published in	Time
[16]	Entire Space Multi-Task Model: An Effective Approach for Estimation Post-Click Conversion Rate | ESMM	SIGIR	2018
[17]	Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts｜MMOE	KDD	2018

 

Others

	Paper	Published in	Time
[1]	Neural Collaborative Filtering vs. Matrix Factorization Revisited	RecSys	2020

树模型，XGB、LGB：

	Paper	Published in	Time
[2]	XGBoost: A Scalable Tree Boosting System	KDD	2016
[3]	LightGBM: A Highly Efficient Gradient Boosting Decision Tree	NIPS	2017

Capsules：

	Paper	Published in	Time
[4]	Dynamic Routing Between Capsules	NIPS	2017

 

Contact Details

作者有一个自己的公众号：推荐算法的小齿轮，如果喜欢里面的内容，不妨点个关注。