推荐系统中的前沿技术研究与落地

1 .!"#$%&'()* +,-./ !"#$%AutoML&'(#$ )* +,-./01 234 

2 .!" • !"#$%&'()*+,-./0 • AutoML56789:;<=&>? • '(#$56789:;<=&>? 

3 .#$%&'()*+,- !"#$ ! !"#$% Matching %&'( ! )*+,- &' Ranking #$ ……. !" (User) (Item) ./01 (&' Re-Ranking 

4 . #$./01234./56 " 2006~23456 (Collaborative Filtering) • Nearest neighbor • Matrix factorization (MF) • Topic models • !"#$%&!"#$%&''()%*+%&),-()* " 2010~2789(:; (Generalized Linear Model) +,-./0123456728 • Logistic regression • : Factorization Machines(FM)/Field-aware FM(FFM) • ./#$9:;<=>?@A'BCDEFG@3-.1 • Learning to rank: BPR, RankSVM, LambdaRank 2'HIJKLMN!OPQ8 " 2015~2<=>?@A:; • 012345RST-./0U3VWXY!Z['\]67 • FNN, PNN, wide & deep, DIN, DeepFM, etc. 89:;<=^!"#$>?@ABCDEF_`abN cd@A8 " 2018~2BC>?@A:; • Multi-armed bandit(MAB), Markov Decision Process (MDP) " 2019~2AutoML@A:; • AutoCross, Neural Input Search (NIS) 

5 .789:#$;<=>?@ABCDE+FGHI … 0.18 … 0.05 … 500 500M 0.02 0 0 1 0 0 1 … 0 1M CTR! 1. Embedding 2. Interaction GHIJKLMNCO 

6 .789:#$;<JK?Embedding + MLP MLP Embedding Embedding !"#$#%&'())'*+&,-./'01234 Deep neural networks for YouTube recommendation (Recsys2016,google) 

7 .MLP'LM+N r r r 1*100 2*100 m*100 • • Embedding size + Latent Cross: Making Use of Context in Recurrent Recommender Systems,wsdm2018 

8 .DEOPQR?embedding + interaction + MLP w0 wi v1i v2i v3i wi v1j v2j v3j … … 1 1 field j field i Product operation CNN Layer LSTM\GRU • !"#$%&'()*+,- • ./012%34+,56 78%9:;.- Attentionef Memory-based network 

9 . DEOPQR?Product and Attention Product Operation Attention Operation 

10 . DEOPQR?RNN/CNN family and Memory-based RNN/CNN family Memory-based 

11 .DeepFM 012#cdc"##$c&e2#f<ghij • kl./012#345mnop*+6qrst=>;?Vuvnw <xy=>z{ |}~<€-‚ƒ#„c Gƒ†‡ff#2cˆ$$‰ E$„#ƒƒ1eGcˆ$$<z{Š@ • ‹ŒŽ)%&*+012#cdc"##$c&e2#f~<op*+‘!’ “”•–~—˜™rš›@ "##$%&' • ()%&*+,-./012#3456789:;<=>;?@ • %&*AB"##$*ACD#EF#221GHI<JK@ • LEF#221GHI<MNOP%&*AB"##$*AQRSTU V<W45XYVZ[6(\#EF#221GH]^\_`ab@ DeepFM: A Factorization-Machine based Neural Network for CTR Prediction (IJCAI 2017,Huawei) 

12 .PIN Prediction Fully Connected Layers Hidden State Sub-net 1 Sub-net 2 Sub-net i FC layer F1 F2 F1*F2 Embedding Embed 1 Embed 2 Embed N Layer Feature 1 Feature 2 Feature N Product-based Neural Network for User Response Prediction over Multi-field Categorical Data(TOIS 2019,Huawei) 

13 .!" • !"#$56789:;<=&>? • AutoML%&'()*+,-./0 • '(#$56789:;<=&>? 

14 .AutoMLS789:#$;<T'UV Embedding • !"#$PQ%&RS>?T!"#$%%&'()*)+,-. *)/01 • RU012345VWXY:;/M!"#$>?@AZ[\]XY^,M:;#$Z_`]abcdefT • !"#$%%&'()2&3$4))56+,-.789/01:; • XY!"#$CDMEF]ghi/j*]kBIlmnJjo 

15 .AutoCross • 0123pq3rr'ghij^kllmnop'q!"#$%&#'()*+&#",+)(,$--'./)*$,)0&1"2&,)3&#&)'.)4+&256$,27)!882'(&#'$.-r • %*stcd67du'vw+vxsotuvwxuyz{|}'mt~€‚ƒ*„ 8†tBC‡ˆj‰Š'‹t~ŒŽ* ‘’“”•–“—!3|}˜A™š0e›+œž)*œ8 • AutoCross: Automatic Feature Crossing for Tabular Data in Real-World Applications(KDD 2019, ) 

16 .NIS • yz{'Ÿ ¡¢£^mnop–¤–on'qy|1q}~€‚12r|}qƒ„ 3q~}q†|rƒ}~|q|ƒ3‡‡|ˆ}2€3‡3ˆ|~rr • !"#$CD‰‡Š|ˆˆ€†r€‹|M:;/@As • -./0Ucd3|}¥¦§¨©ª«£¬¬­®¡¯‘"~|}'Œ3ƒ}Š1~}qr€‹|=‰‡Š|ˆˆ€†r€‹|(°¦3±²³' ´µ3|}¥¦¶·‘3¸¹8 • ŽM‘’“”•M–—/'º<»¼3Œ3ƒ}Š1~}qr€‹|=‰‡Š|ˆˆ€†r€‹|½¾-.œ3¿ÀÁ»u8 • • • ENAS • vÂsotãÄÅ¢¢ÆoÇÈɖÊË8mtÌÍÎÇÈo–ÊË Neural Input Search for Large Scale Recommendation Models(2019,Google) 

17 .789:,WT'AutoMLXY =&:+,)-';+ 2*3 ($..+(#'$.- *+&#",+)($%1'.&#'$.) .'/$-+0/&1')*+,$- -+2+(#'$.<8,".'./ 9$(&1"2&,:)-';+ !"#$%%&'()*+,$- +%1+77'./)-';+ • DEFGHIJHKLMNOPLQQHRSMTDNU>? • VWXY>?2XYVW56Z[\VW]^ • _`abcdeZfg • 456789:;<=>?//$'/&1'@ABB@CBB@3-1%D0/EF 

18 .AutoGroup:OPDEZ[\]^ hijkl4\VWXYcmnVWopq 

19 .AutoGate:DEOP'Z[\_` • 

20 .!" • !"#$56789:;<=&>? • AutoML56789:;<=&>? • 12#$%&'()*+,-./0 

21 .abc\9:'#$;< rstu>?c@Avwxcyz{| AlphaGo • !"#$% • &'$% 15% • $('% (Reinforcement Learning) • • + ! 

22 . abc\9:'#$;< policy-based value-based policy & value-based Reinforcement learning in large discrete action space. In Adapting markov decision process for search result Recommendations with negative feedback via pairwise 2015 . diversification. SIGIR 2017 deep reinforcement learning. KDD 2018 Deep reinforcement learning for page-wise recommendation. Top-K off-policy correction for a REINFORCE DRN: A deep reinforcement learning framework for news RecSys 2018 . Recommender system. WSDM 2019 . recommendation. WWW 2018 Reinforcement learning to rank in e-commerce search engine: formalization, analysis, and application. KDD 2018 . 

23 .YouTube?c\9:;<defghi!'jklm • Top-K Off-Policy Correction for a REINFORCE • Reinforcement Learning for Slate-based Recommender Systems: A Tractable Decomposition and Practical Methodology zžp0zš›œ Recommender System, ˜{™4š›œ • • YouTube item LTV long- • REINFORCE term value • top-K off-policy • CTR LTV • • 

24 .nopq?rc\9:KsOPtuvwxy • • • • • Virtual Taobao: Virtualizing Real-World Online Retail Environment for Reinforcement Learning (AAAI 2019, 

25 .TPGR • • • • Large-scale Interactive Recommendation with Tree-structured Policy Gradient (AAAI 2019, 

26 .z{|}~ • <=>?vw/<=>?@A}~v€vw‚ƒB34+,56*CDE%FGHIE- • BC>?JKLMNOPQRA!"STUVQ%56DEAWXYZ[\]/^_`- • „EG D†abc56*Cd%efgh/ijAklVWophi‡ˆm<=>?‰bŠ^Z:;fgno pq]rsQR- 

27 . Publication List Year 2017: • Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, Xiuqiang He: DeepFM: A Factorization-Machine based Neural Network for CTR Prediction. IJCAI 2017 • Huifeng Guo, Ruiming Tang, Yunming Ye, Xiuqiang He: Holistic Neural Network for CTR Prediction. WWW 2017 • Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, Xiuqiang He: A Graph-Based Push Service Platform. DASFAA 2017 Year 2018: • Weiwen Liu, Ruiming Tang, Jiajin Li, Jinkai Yu, Huifeng Guo, Xiuqiang He, Shengyu Zhang: Field-aware Probabilistic Embedding Neural Network for CTR Prediction. RecSys 2018 • Feng Liu, Ruiming Tang, Xutao Li, Yunming Ye, Huifeng Guo, Xiuqiang He: Novel Approaches to Accelerating the Convergence Rate of Markov Decision Process for Search Result Diversification. DASFAA 2018 Year 2019: • Haokun Chen, Xinyi Dai, Han Cai, Weinan Zhang, Xuejian Wang, Ruiming Tang, Yuzhou Zhang, Yong Yu: Large-scale Interactive Recommendation with Tree- structured Policy Gradient. AAAI 2019 • Yanru Qu, Bohui Fang, Weinan Zhang, Ruiming Tang, Minzhe Niu, Huifeng Guo, Yong Yu, Xiuqiang He: Product-based Neural Network for User Response Prediction over Multi-field Categorical Data. TOIS 2019 • Bin Liu, Ruiming Tang, Yingzhi Chen, Jinkai Yu, Huifeng Guo, Yuzhou Zhang: Feature Generation by Convolutional Neural Network for Click-Through Rate Prediction. WWW 2019 • Huifeng Guo, Ruiming Tang, Yunming Ye, Feng Liu, Yuzhou Zhang: A Novel Approach for Session-based Recommendation. PAKDD 2019 • Wei Guo, Ruiming Tang, Huifeng Guo, Jianhua Han, Wen Yang, Yuzhou Zhang: Order-aware Embedding Neural Network for CTR Prediction. SIGIR 2019 • Huifeng Guo, Jinkai Yu, Qing Liu, Ruiming, Yuzhou Zhang: PAL: A Position-bias Aware Learning Framework for CTR Prediction in Live Recommender Systems. RecSys 2019 • Jianing Sun, Yingxue Zhang, Chen Ma, Mark Coates, Huifeng Guo, Ruiming Tang, Xiuqiang He: Multi-Graph Convolution Collaborative Filtering. ICDM 2019 

28 .tangruiming@huawei.com