WWW 2019 Digest

WWW2019 論文読み会

Transcript

1. WWW 2019 
    
      WWW 2019  2019/09/02
2. LINE Data Labs Data Science 1 Team LINE • AB • • •
   
   
   e 2018d 5 2d 2018d8 LINE
3. 
    
    • #"#021 • %,) 
    ! 
   5. • #"#(&
   /4  * • The Web Conference 20196$'3+-
4. Overview
5. Overview
   
    4%3!" 9.5 ;< 6
   WWW#$!" -    1?Web 43'27* +)8,   DAY RESEARCH TRACK DAY RESEARCH TRACK DAY RESEARCH TRACK 5/15 Fairness, Credibility and Search 5/15 Efficiency and Scalability 5/16 Health on the Web 5/15 Network Algorithms 5/15 Search 5/16 Sarcasm, Sentiment, and Language 5/15 Recommendation 5/16 Networks, Opinions, and Perceptions 5/17 Graph Models 5/15 Security 5/16 Personalization! 5/17 Sarcasm, Sentiment, and Language 5/15 Knowledge Synthesis 5/16 Crowdsourcing and Human Computation 5/17 Economics, Monetization, and Online Markets 5/15 Knowledge Analysis and Querying 5/16 Text Classification and Relation Extraction 5/17 Social Recommendation and Experimentation 5/15 Communities, Complaints, and Collective Action 5/16 Systems and Infrastructure 5/17 Topic Modeling and Representation 5/15 Network Applications 5/16 Network and Analysis 5/15 Privacy and Trust 5/16 Mobile and Ubiquitous Computing Research Track/:0<&  = > (@ 
6. Overview
     0>HD8
   "#?/100$70> 
   A4 @15;,*F
   !9& =(MicrosoftGoogleAlibaba GroupStanford=(G-=(E+.BC<  
    #<4'  WWW
   Web
   "  !   2:)B%63 
7. Day1-Day2 1. WorkshopInternational Working on Modeling Social Media 2. TutorialA/B Testing at Scale (Microsoft) 3. TutorialOnline User Engagement (Spotify)! 4. BIGRecommending and Searching, Research @ Spotify 5. TutorialPrivacy-preserving Data Mining in Industry (LinkedIn) Day3-Day5 • Keynote, Poster Session • Research Track     
   
       Sponsor Booth Day1-Day223 Workshop
   22 Tutorial Day3-Day512 Research Track
   Poster Session
   Keynote
8. Tutorial
9. Tutorial: Online User Engagement Slides
   h(ps://onlineuserengagement.github.io/ Speaker: Mounia Lalmas Research Director & Head of Tech Research @Spo?fy 20HKUser Engagement Personaliza?on Track Chair BIG Track Plenary Speaker Topic 1.Metrics • '%'$
   
   3E?B7F@1C =N • .7F  96 #& 8OQI)M ->.30(; QI)M User Engagement /PQIA<4 L* %! 2.Op<miza<on • A/B "' 5D+ • WSDM2019 QI :, Spo?fy 8OJG :,
10. '"?'# FK$)(!!(%*A3VU .Sleep$)(!!(%*A3VU -> Sleep$)(!Dwell time O 9=  ->!(%*A3VU&$* * $M -L ! Tutorial: Online User Engagement Slides
    https://onlineuserengagement.github.io/ Paolo Dragone, Rishabh Mehrotra and Mounia Lalmas. Deriving User- and Content-specific Rewards for Contextual Bandits. WWW 2019. Metrics
    Dwell Ame as streaming Ame SpoDfyContextual BanditWC4GI: • DZ!(%*A32N/30T,F@H • Q+BX&* *86>\' (*
    'R!(%*A3VU 5
    1'<YWCP7 • ;EDZS2 25,F@H[0J"
11. Research Track Personaliza1on
12. 
      Exploring Perceived Emotional Intelligence of Personality-Driven Virtual Agents in Handling User Challenges O; 5?=L`D 8T 3'(VA) Emotional IntelligenceRC Listening Between the Lines: Learning Personal Attributes from Conversations :h%!'MB
    ]g
    9W EOVQPY Dual Neural Personalized Ranking /U %,UfcPairwise033d 033<IZ4 Dynamic Ensemble of Contextual Bandits to Satisfy Users' Changing Interests! /@Nb7 ^_YK[ >A X6 3%!#.2*3&$'21",Z4 Quality Effects on User Preferences and Behaviors in Mobile News Streaming Je/F\+(*$ ).!a HSG -&2Z4 Research Track - Personalization Day4 10:30-12:30 / Chair: Mounia Lalmas
13. Research Track - Personalization: Dynamic Ensemble of Contextual Bandits to Satisfy User’s Changing Interests
14.  • • Ar#cle1, Ar#cle4 Ar#cle2 t3 Ar#cle3 t1,t2,t4 
    • γ -Restart algorithm discounted-UCB algorithm >> discount factor • Cao >> 1 • Wu 2 contextual bandit model 1 1 >> 1 context Research Track - Personalization: Dynamic Ensemble of Contextual Bandits to Satisfy User’s Changing Interests admissible
15. . 3 132! e n mu p n bs i o . 3 3 1" n a i o n a ld t 1 3 . 22 132# r n a i o e x r Expected Reward !" $ % & ' $ & % ' ( ⛹ * + ( ⛹ * + cn A S B Time Research Track - Personaliza2on: Dynamic Ensemble of Contextual Bandits to Sa2sfy User’s Changing Interests #$% ∗ #$%'( ∗ )"( ( ∆+ !" ≠ 0 #$% ∗ #$%'( ∗ )". $ ∆+ !" = 0 feture vector : )"0 user preference vector : #∗ + !" = )" 1# ∗ )"( , )". , … )"4
16. 9,F[p}dw?j>"*<2 )-08';5<3(14i= & aMKHo   ~v)7  Gep}dEIf>"5<3(14Sz 
    *<2)-4Yy%D|  Sz
    t]5<3(1473;UN% xP" 
    hXZgyTus%\hO " " t]5<3(1473;UN"Bandit expert
    expert\hrQ%q@"Bandit auditor%lm';+:.6     
        "  {LJW  N_3/3R3/JWRegret%oC  $
    DenBand `Bw?k%JV  c f>CTR\h^n# baseline !Ab# Research Track - Personaliza2on: Dynamic Ensemble of Contextual Bandits to Sa2sfy User’s Changing Interests
17. Q.{1FKc+@7)?@0UD&-:4gvl A.gv(‡† Qfux AC • pˆ‚(]}  eˆ‚,=21*28 RF 
    L%`suˆ‚(z![H(j&   • InM„(*2834&koX m & • rO …|~ "N|ubE & • \^AG-60‰#fbE <19.>@d $&(y  • WZ€(T&
    tJuiVhSdBƒaP _'&wY&!; 5/=q 2020TAIPEI WWW 2019 UD?:4 h/ps://engineering.linecorp.com/ja/blog/the-web-conference-2019-report/
18. Bronze 
    
     The Web Conference 
     
     
      
19. ,- Exploratorium +'!-
    "%
    ,-& ) $ (#  / *. 0  The Web Conference  
       
20. The Web Conference  
        Best Poster (5681 9
    :?< + '2Web4!&$*) #"%03   ,-7;.> "=/   h,ps://twi,er.com/zijianwang30/status/1129130581806047232
21. Tutorial • Online User Engagement: Metrics and Op4miza4on • h8ps://onlineuserengagement.github.io/ Reference Paper • Deriving User- and Content-specific Rewards for Contextual Bandits • h8ps://labtomarket.files.wordpress.com/2019/03/www2019_re wards.pdf • Dynamic Ensemble of Contextual Bandits to Sa4sfy Users' Changing Interests • h8p://www.cs.virginia.edu/~hw5x/paper/WWW2019- DenBandit-Wu.pdf • Exploring Perceived Emo4onal Intelligence of Personality-Driven Virtual Agents in Handling User Challenges • h8ps://dl.acm.org/cita4on.cfm?id=3313400 • Listening Between the Lines: Learning Personal A8ributes from Conversa4ons • h8ps://arxiv.org/abs/1904.10887 • Dual Neural Personalized Ranking • h8ps://dl.acm.org/cita4on.cfm?id=3313585 • Quality Effects on User Preferences and Behaviors in Mobile News Streaming • h8p://www.thuir.org/group/~YQLiu/publica4ons/WWW2019Lu.pdf
22. Appendix
23. Dynamic Ensemble of Contextual Bandits ) • b !" o #$ • b o / Ø admissibleFbandit expert d c Ø admissibleDF id BFexpert auditor m Expected Reward T . 1 1,2 ls t ( . T . 1 1,2 rDe B n ( . 2-, %&'$," = 1 +" $ , -∈/0 1 %&'$," (3) 5 3$," = arg 39:-∈/0 1 ( ̂ <$," 3 − ' $," > ) !" = arg 3!?$∈@ %&'$," ̂ <$," 3 ≤ lsvg a =Badness a ! 3B ! 3C " # $ % " $ # % & ⛹ ( ) & ⛹ ( ) " 3B admissible & admissibleDF expert Time Research Track - Personaliza>on: Dynamic Ensemble of Contextual Bandits to Sa>sfy User’s Changing Interests
24. Research Track -Personaliza2on: Dynamic Ensemble of Contextual Bandits to Sa2sfy User’s Changing Interests Dynamic Ensemble of Bandit Experts ; DenBand
     o B M A is C Bb d c b pD • 3( - R v h W L M g nF U • 3 3( - S g L TT t B ( ) M A is • ) - R e Y a Wr L M • 3 1 3 L 1 l L