Upgrade to Pro — share decks privately, control downloads, hide ads and more …

オンライン広告におけるCTR/CVR推定関係の論文を30本くらい雑に紹介する / rtb-papers-ctr

todesking
December 23, 2019

オンライン広告におけるCTR/CVR推定関係の論文を30本くらい雑に紹介する / rtb-papers-ctr

todesking

December 23, 2019
Tweet

More Decks by todesking

Other Decks in Technology

Transcript

  1. ΦϯϥΠϯ޿ࠂʹ͓͚Δ
    CTR/CVRਪఆؔ܎ͷ࿦จΛ
    30ຊ͘Β͍ࡶʹ঺հ͢Δ
    rtb-papersฤ
    @todesking

    View full-size slide

  2. rtb-papersͱ͸
    • ্ւަ௨େֶͷWeinan ZHANGॿڭत͕·ͱΊ͍ͯΔɺ
    ΦϯϥΠϯ޿ࠂؔ܎ͷ࿦จू

    • https://github.com/wnzhang/rtb-papers

    • δϟϯϧ͝ͱʹओཁͳ࿦จ͕·ͱ·͍ͬͯΔͷͰɺ޿ࠂܥ
    ͷػցֶशٕज़ͷೖ໳ʹ͓͢͢Ί

    • ࠓճ͸CTR/CVRਪఆʹؔ͢Δ࿦จΛ঺հ͠·͢

    View full-size slide

  3. CTR/CVRਪఆ
    • Click Through Rate/Conversion Rate

    • ޿ࠂΛݟͨͱ͖ͷϢʔβͷԠ౴Λ༧ଌ͢ΔͷͰɺ·ͱΊͯ
    user response predictionͱ͔ݺ͹ΕΔ

    • ޿ࠂͷΫϦοΫ/ίϯόʔδϣϯ(঎඼ͷߪೖͳͲɺͦͷ޿
    ࠂ͕ୡ੒͍ͨ͠໨త)Λ༧ଌ͢ΔλεΫ

    • యܕྫ: WebϖʔδpʹϢʔβu͕དྷ๚ͨ͠ɻ޿ࠂaΛදࣔ
    ͨ͠৔߹ɺΫϦοΫ/ίϯόʔδϣϯ͕ൃੜ͢Δ֬཰͸͍
    ͘Β͔

    • ಛ௃ྔ͸ΧςΰϦΧϧσʔλ͕ଟ͍

    View full-size slide

  4. CTR/CVRਪఆ
    • ҰൠతʹɺCVRਪఆͷ΄͏͕೉͍͠

    • ਖ਼ྫ͕গͳ͍

    • ΫϦοΫ཰Ͱ΋~1%ఔ౓ɺCVR͸͞Βʹ௿͍

    • CVͷఆٛ͸ଟ༷

    • ಛఆϖʔδ΁ͷΞΫηε͔Β঎඼ͷߪೖ·Ͱ༷ʑ

    • ΞτϦϏϡʔγϣϯϞσϧ

    • ෳ਺ͷimp/clickͷޙCVͨ͠৔߹ɺͲͷimpͷ੒Ռʹ͢Δ͔

    • ͍Ζ͍ΖͳϞσϧ͕͋Δ(࠷ޙʹΫϦοΫ͞Εͨ΋ͷΛ࠾༻ɺෳ਺ͷ
    impʹ੒ՌΛ෼഑)

    • ஗ԆϑΟʔυόοΫ

    • imp ʙ click͸͍͍ͤͥ਺෼͕ͩɺCV͢Δʹ͸΋ͬͱ௕͍͕͔͔࣌ؒΔ

    • ʮෛྫʯͱʮCVσʔλະ౸ணʯͷݟ෼͚͕͔ͭͳ͍

    View full-size slide

  5. Evaluating and Optimizing Online Advertising:
    Forget the click, but there are good proxies by
    Brian Dalessandro et al. SSRN 2012.
    • "ΦϯϥΠϯ޿ࠂͷධՁͱ࠷దԽ: ΫϦοΫΑΓྑ͍୅ସࢦඪ"

    • ͋Δϒϥϯυͷ޿ࠂΛݟͨਓ͕7೔Ҏ಺ʹͦͷϒϥϯυͷ঎඼Λങ
    ͏͔Λ༧ଌ͢Δέʔε

    • CVσʔλ͸εύʔεա͗ͯେม

    • ؆୯ʹखʹೖͬͯɺCVͱ૬͕ؔ͋ΔΑ͏ͳ୅ସࢦඪ(proxy)͕͋Δ
    ͱخ͍͠

    • ௚ײతʹ͸ΫϦοΫ͕୅ସͱͳΓͦ͏͕ͩɺ࣮͸΄΅ؔ܎ͳ͍

    • ΫϦΤΠςΟϒͷσβΠϯ͕ΫϦοΫ࠷దԽΛ໨తͱ͍ͯ͠Δ
    ͔ΒͰ͸?

    • ͦͷ୅ΘΓʹɺϒϥϯυαΠτ΁ͷ๚໰͸Α͍୅ସࢦඪͱͳΔ

    View full-size slide

  6. Estimating Conversion Rate in Display
    Advertising from Past Performance
    Data by Kuang-chih Lee et al. KDD 2012.
    • "աڈͷύϑΥʔϚϯεσʔλ͔ΒͷCVRਪఆ"

    • ϢʔβɺWebϖʔδɺ޿ࠂͷ૊Έ߹Θ͔ͤΒCVRΛ༧ଌ͢
    Δ

    • ͦΕͧΕΛ֊૚తʹ෼ׂͯ͠աڈͷύϑΥʔϚϯεΛूܭ

    • ϩδεςΟοΫճؼͰͦΕΒΛ౷߹ͯ͠CVR༧ଌ

    View full-size slide

  7. Ad Click Prediction: a View from the
    Trenches by H. Brendan McMahan. KDD
    2013.
    • "ΫϦοΫਪఆ: ᆠߺ͔ΒͷோΊ"

    • Google

    • େن໛ͳCTR༧ଌϞσϧͷӡ༻͔ΒಘΒΕͨ஌ݟͷ঺հ

    • ϩδεςΟοΫճؼͷΦϯϥΠϯֶश(FTRL-Proximal, Per-
    coordinate learning rates)

    • ϝϞϦઅ໿ͷٕ๏

    • ϞσϧධՁख๏

    • ༧ଌͷ৴པ౓Λද͢uncertainty score

    • ༧ଌͷΩϟϦϒϨʔγϣϯ

    • ಛ௃ྔͷ؅ཧ

    • ͏·͍͔͘ͳ͔ٕͬͨ๏

    View full-size slide

  8. Practical Lessons from Predicting Clicks
    on Ads at Facebook by Xinran He et al.
    ADKDD 2014.
    • "Facebookʹ͓͚ΔΫϦοΫ༧ଌ͔Βͷڭ܇"

    • GBDTͷ݁ՌΛઢܗճؼͰ࢖͏stackingϞσϧ

    • ઢܗճؼ͸LRͱBOPR

    • BOPR: Bayesian online learning scheme for probit
    regression

    • σʔλ઱౓͕ॏཁͳͷͰΦϯϥΠϯֶश͢Δ

    • Online joiner: ϩάσʔλΛΦϯϥΠϯֶशػʹ৯Θͨ͢
    ΊͷγεςϜ

    • ϞσϧαΠζͷ࡟ݮ

    • ֶशσʔλͷμ΢ϯαϯϓϦϯά

    View full-size slide

  9. Modeling Delayed Feedback in Display
    Advertising by Olivier Chapelle. KDD 2014.
    • "σΟεϓϨΠ޿ࠂʹ͓͚Δ஗ԆϑΟʔυόοΫϞσϦϯ
    ά"

    • Criteo

    • CV͸impޙ௕࣌ؒܦ͔ͬͯΒൃੜ͢Δ

    • ͜ͷ஗ԆΛߟྀͨ͠CV༧ଌϞσϧΛ࡞ͬͨ

    View full-size slide

  10. Scalable Hands-Free Transfer Learning for
    Online Advertising by Brian Dalessandro
    et al. KDD 2014.
    • "ΦϯϥΠϯ޿ࠂʹ͓͚Δεέʔϥϒϧͳख͍ؒΒͣసҠֶश"

    • DistilleryͷCVR༧ଌγεςϜ

    • Ωϟϯϖʔϯ͝ͱʹϞσϧ͕͋Δ

    • ίʔϧυελʔτ໰୊ɺCVͷଟ༷ੑɺϝϯςͷखؒͳͲͷ໰
    ୊ʹରॲ

    • Bayesian transfer learning: ࣄલ෼෍ʹΑΓίʔϧυελʔτ
    Λ؇࿨

    • ϋΠύʔύϥϝʔλෆཁͳSGD LR(NoPesky learning rates +
    Adaptive regularization)

    View full-size slide

  11. Scalable Hierarchical Multitask Learning
    Algorithms for Conversion Optimization in Display
    Advertising by Amr Ahmed et al. WSDM 2014.
    • "CV࠷దԽͷͨΊͷεέʔϥϒϧͳ֊૚ϚϧνλεΫֶश"

    • Google

    • ؔ࿈ͨ͠໰୊͸·ͱΊͯղ͍ͨ΄͏͕ੑೳ͕ྑ͍

    • CV(attributed/non-attributed)ɺΫϦοΫΛҰ౓ʹ༧ଌ͢Δ
    Ϟσϧ

    • Multitask-learning

    • ֊૚తͳ֬཰ϞσϧͰΩϟϯϖʔϯɾλεΫؒͷґଘΛදݱ

    • ֊૚ϚϧνλεΫֶशΛ෼ࢄॲཧͰεέʔϧͤ͞Δ࿩

    View full-size slide

  12. Predicting Response in Mobile Advertising with
    Hierarchical Importance-Aware Factorization
    Machine by Richard J. Oentaryo et al. WSDM 2014.
    • "֊૚తImportance-aware FMʹΑΔϞόΠϧ޿ࠂͷϨεϙϯε
    ༧ଌ"

    • Λਪఆ͍ͨ͠(Ͳ͏͍͏ঢ়گઃఆ??)

    • page, adʹ͸֊૚ߏ଄͕͋Δ

    • ֊૚ͷ֤ϊʔυʹରԠ͢ΔॏΈΛਖ਼ଇԽͰ੍ޚ(ಉ͡਌ʹଐ͢
    ΔϊʔυͷॏΈ͸͍ۙ)

    • ֊૚͝ͱʹू໿ֶͨ͠शσʔλΛ࡞Γɺαϯϓϧͷଟ͞Λॏཁ
    ౓ͱͯ͠ॏΈ෇ֶ͚ͯ͠श

    • Ͳ͔ͬͰtemporal dynamicsΛߟྀͯ͠ΔΒ͍͕͠ݟࣦͬͨ
    P(click|page, ad, day)

    View full-size slide

  13. Simple and Scalable Response Prediction for
    Display Advertising by Olivier Chapelle Criteo,
    Eren Manavoglu, Romer Rosales. ACM TIST 2014.
    • "σΟεϓϨΠ޿ࠂʹ͓͚ΔγϯϓϧͰεέʔϥϒϧͳԠ౴༧ଌ"

    • Criteo+MS+LinkedIn

    • LRͰCTR/CVR͢Δ࿩

    • Bayesian LRͰThompson sampling͢Δ࿩୊΋

    View full-size slide

  14. Offline Evaluation of Response Prediction
    in Online Advertising Auctions by Olivier
    Chapelle. WWW 2015.
    • "ΦϯϥΠϯ޿ࠂΦʔΫγϣϯʹ͓͚ΔԠ౴༧ଌͷΦϑϥ
    ΠϯධՁ"

    • Criteo

    • CTRਪఆͷධՁͱͯ͠͸Ұൠతͳڭࢣ෇ֶ͖शͷࢦඪ͕࢖
    ΘΕ͍ͯΔ͕ɺਪఆ݁Ռ͕Ͳ͏࢖ΘΕΔͷ͔ߟ͑Δ΂͖

    • ਪఆͨ͠CTRʹΑͬͯbidՁ֨ΛܾΊΔγφϦΦʹ͓͚Δ
    ධՁࢦඪͱͯ͠expected utility[5]ΛఏҊɺ࣮σʔλͰݕূ

    View full-size slide

  15. Deep Learning over Multi-field Categorical Data: A
    Case Study on User Response Prediction by
    Weinan Zhang, Tianming Du, Jun Wang. ECIR 2016.
    • "ϚϧνϑΟʔϧυΧςΰϦΧϧσʔ
    λͷਂ૚ֶश: ϢʔβԠ౴ਪఆͷࣄྫ"

    • ը૾΍Ի੠ͱҟͳΓɺ޿ࠂͰ࢖͏
    σʔλ͸ෳ਺ϑΟʔϧυͷΧςΰϦΧ
    ϧ஋

    • ͜ΕΒͷަޓ࡞༻ͷ͏ͪ༗ޮͳ΋ͷ
    ΛࣗಈͰൃݟ͍ͨ͠

    • Factorization Machine supported NN
    ͓Αͼ Sampling-based NNͱ͍͏Ϟ
    σϧΛఏҊ

    View full-size slide

  16. Factorization Machines with Follow-The-
    Regularized-Leader for CTR prediction in Display
    Advertising by Anh-Phuong Ta. BigData 2015.
    • "FMͱFTRLʹΑΔCTRਪఆ"

    • ΦϯϥΠϯ࠷దԽΞϧΰϦζϜͰ͋ΔFTRL-ProximalΛ
    FMʹద༻ͨ͠FTRFL(Follow-The-Regularized-Factorized-
    Leader)ͷఏҊ

    • FMͷΦϯϥΠϯֶश͕ߴ଎+ߴੑೳʹͳͬͨ

    View full-size slide

  17. A Convolutional Click Prediction Model by
    Qiang Liu, Feng Yu, Shu Wu, Liang Wang.
    CIKM 2015.
    • "৞ΈࠐΈΫϦοΫ༧ଌϞσϧ"

    • imp୯ମͰ͸ͳ͘ܥྻΛ࢖ͬͯCTR༧ଌ͢ΔϞσϧ

    • impͷܥྻʹରԠ͢Δಛ௃ྔΛೖྗɺ৞ΈࠐΈ͢Δ

    View full-size slide

  18. Cost-sensitive Learning for Utility Optimization in Online
    Advertising Auctions by Flavian Vasile, Damien Lefortier,
    Olivier Chapelle. Extension under-review of the paper
    presented at the Workshop on E-Commerce, NIPS 2015.
    • "ίετΛߟֶྀͨ͠शʹΑΔΦϯϥΠϯ޿ࠂͷutility࠷దԽ"

    • Criteo+FB+Google

    • CTRਪఆͷੑೳ͸utilityͰଌΔͷ͕࣮৘ʹ߹͍ͬͯΔ

    • ͩͬͨΒ࠷ॳ͔ΒutilityΛ࠷దԽ͍ͨ͠

    • ॏΈ෇͖log lossΛ࠷దԽ͢Δ͜ͱͰۙࣅతʹutilityΛ࠷దԽ
    Ͱ͖Δ

    • ղઆ: https://k11i.biz/blog/2017/08/15/2017-adkdd-criteo/

    View full-size slide

  19. User Response Learning for Directly Optimizing
    Campaign Performance in Display Advertising by Kan
    Ren, Weinan Zhang, Yifei Rong, Haifeng Zhang, Yong Yu,
    Jun Wang. CIKM 2016.
    • "ΩϟϯϖʔϯͷύϑΥʔϚϯεΛ௚઀࠷దԽ͢ΔͨΊͷ
    ϢʔβԠ౴ͷֶश"

    • CTR༧ଌϞσϧ͸޿ࠂ഑৴γεςϜͷҰ෦Ͱ͋Δ

    • Ϟσϧ୯ମΛ࠷దԽ͢ΔΑΓɺγεςϜશମΛ࠷దԽͨ͠
    ͍

    • LRʹΑΔCTR༧ଌΛؚΉ഑৴γεςϜΛϞσϧԽ͠ɺ
    ΩϟϯϖʔϯͷརӹΛ௚઀࠷దԽͨ͠

    • A/BςετͰ΋޷੒੷

    View full-size slide

  20. Predicting ad click-through rates via feature-based fully
    coupled interaction tensor factorization by Lili Shan, Lei
    Lin, Chengjie Sun, Xiaolong Wang. Electronic Commerce
    Research and Applications 2016.
    • "Feature-based fully coupled interaction tensor factorizationʹΑ
    ΔCTR༧ଌ"

    • (Ϣʔβ, Webϖʔδ, ޿ࠂ)Λςϯιϧͱͯ͠දݱɺCTRਪఆΛς
    ϯιϧิ׬໰୊ͱͯ͠ղ͘

    • Fully-Coupled interactions Tensor Factorizationͱ͍͏ΞϧΰϦζ
    ϜΛఏҊ

    View full-size slide

  21. Bid-aware Gradient Descent for Unbiased Learning
    with Censored Data in Display Advertising by Weinan
    Zhang, Tianxiong Zhou, Jun Wang, Jian Xu. KDD 2016.
    • "ܽଛͷ͋ΔσʔλͷͨΊͷೖࡳߟྀܕޯ഑߱Լ๏ʹΑΔ
    όΠΞεͳֶ͠श"

    • RTB޿ࠂʹ͓͍ͯɺCTRਪఆʹ࢖ΘΕΔσʔλ͔Β͸ʮམ
    ࡳͰ͖ͳ͔ͬͨimpʯ͕ܽམ͍ͯ͠Δ(Censored data)

    • ੜଘ෼ੳΛద༻ͯ͠ɺ͜ΕΒͷܽଛΛߟྀͨ͠SGDΛՄ
    ೳʹ͢Δख๏ͷఏҊ

    View full-size slide

  22. Deep CTR Prediction in Display
    Advertising by Junxuan Chen et al. MM
    2016.
    • "ਂ૚CTRਪఆ"

    • Alibaba

    • ඪ४తͳಛ௃ྔʹՃ͑ͯόφʔը૾Λ࢖༻ͨ͠CTRਪఆ

    View full-size slide

  23. Sparse Factorization Machines for Click-
    through Rate Prediction by Zhen Pan et al.
    ICDM 2016.
    • "εύʔεFMʹΑΔCTR༧ଌ"

    • CTRਪఆʹ࢖͏σʔλ͸ඇৗʹεύʔε

    • ͜ͷΑ͏ͳσʔλʹରԠͨ͠Sparse Factorization
    MachinesͷఏҊ

    • ύϥϝʔλͷ෼෍ͱͯ͠ਖ਼ن෼෍ͷ͔ΘΓʹϥϓϥε෼෍
    Λ࢖༻(εύʔεσʔλʹڧ͍)

    • Spark্Ͱͷ෼ࢄֶशΛ࣮૷

    View full-size slide

  24. Product-based Neural Networks for User
    Response Prediction by Yanru Qu et al.
    ICDM 2016.
    • "ProductϕʔεNNʹΑΔϢʔβԠ౴༧ଌ"

    • Product layerʹ͓͍ͯɺ಺ੵ/֎ੵΛ࢖ͬͯಛ௃ྔͷม׵Λ͍ͯ͠Δ

    View full-size slide

  25. Field-aware Factorization Machines in a Real-
    world Online Advertising System by Yuchin Juan,
    Damien Lefortier, Olivier Chapelle. ArXiv 2017.
    • "ΦϯϥΠϯ޿ࠂγεςϜʹ͓͚ΔFFM"

    • ࣮γεςϜͷCTR/CVR༧ଌʹFFMΛಋೖͨ͠

    • ෼ࢄֶश

    • ϞσϧΛ࠶ֶश͢Δͱ͖ʹաڈͷύϥϝʔλΛ࠶ར༻͢Δ
    warm-startingख๏

    View full-size slide

  26. Neural Feature Embedding for User Response
    Prediction in Real-Time Bidding (RTB) by Enno
    Shioji, Masayuki Arai. ArXiv 2017.
    • "ϢʔβԠ౴ਪఆͷͨΊͷχϡʔϥϧຒΊࠐΈ"

    • Ϣʔβͷಛ௃Λදݱͨ͠෼ࢄදݱΛ࡞Γ͍ͨ

    • ϢʔβͷӾཡཤྺΛݩʹɺwebϖʔδΛ୯ޠͱݟͳͯ͠
    CBOWͰϕΫτϧԽ

    • Ϣʔβ͕Ӿཡͨ͠ίϯςϯπͷϕΫτϧΛ౷߹ͯ͠Ϣʔβ
    ͷຒΊࠐΈΛಘΔ(ਖ਼نԽ+ฏۉ)

    • ྨࣅ: Tagami, Yukihiro, Hayato Kobayashi, Shingo Ono
    and Akira Tajima. “Distributed Representations of Web
    Browsing Sequences for Ad Targeting.” (2016).

    View full-size slide

  27. SEM: A Softmax-based Ensemble Model for CTR
    Estimation in Real-Time Bidding Advertising by
    Wen-Yuan Zhu et al. BigComp 2017.
    • "SEM: SoftmaxϕʔεΞϯαϯϒϧϞσϧʹΑΔCTRਪఆ"

    • ಛ௃ྔ͕ΧςΰϦΧϧσʔλͷ৔߹ɺtest࣌ʹॳΊͯग़Δ
    ஋͸ແࢹ͞ΕΔͱ͍͏໰୊

    • Hashing trickͰಛ௃ྔΛϕΫτϧԽˠࣗಈಛ௃ྔબ୒ˠಛ
    ௃ྔ͝ͱʹφΠʔϒϕΠζͯ͠Softmax

    • ৽ن஋໰୊͕hashing trickͰղܾ͢Δͱ͍͏ओு͸Θ͔Β
    ͳ͍Ͱ͢Ͷ……

    • શମతʹṖ

    View full-size slide

  28. Learning Piece-wise Linear Models from Large
    Scale Data for Ad Click Prediction by Kun Gai,
    Xiaoqiang Zhu, Han Li, et al. Arxiv 2017.
    • "CTR༧ଌͷͨΊͷPeace-wiceઢܕϞσϧͷେن໛ֶश"

    • Alibaba

    • 2012೥ҎདྷAlibabaͰ࢖ΘΕ͍ͯΔCTR༧ଌϞσϧ

    • Linear Modelͱݴ͍ͭͭඇઢܗͳݱ৅Λѻ͑Δ

    • ಛ௃ۭؒΛ෼ׂ͠ɺྖҬ͝ͱʹઢܕϞσϧΛద༻͢Δ

    • େن໛σʔλͷֶशʹ଱͑Δ
    σ: dividing function
    η: fitting function

    View full-size slide

  29. DeepFM: A Factorization-Machine based
    Neural Network for CTR Prediction by
    Huifeng Guo et al. IJCAI 2017
    • "DeppFM: CTR༧ଌͷͨΊͷFMϕʔεNN"

    • ಛ௃ΤϯδχΞϦϯάෆཁͱ͔ڧ͍͜ͱ͕ॻ͍ͯ͋Δ

    • Wide and Deepͷൃలܥ

    • ୯ମͷಛ௃ྔ΋ߴ࣍ͷަޓ࡞༻΋େࣄʹ͍ͨ͠ͱ͍͏ؾ࣋ͪ

    • ղઆ: https://data.gunosy.io/entry/deep-factorization-machines-2018

    View full-size slide

  30. Deep Interest Network for Click-Through
    Rate Prediction by Guorui Zhou et al.
    ArXiv 2017.
    • "CTRਪఆͷͨΊͷDeep Interest Network"

    • Alibaba

    • ϢʔβͷߦಈཤྺΛݩʹCTRਪఆ͍ͨ͠

    • ࣌ܥྻσʔλΛݩʹɺϢʔβͷࠓͷঢ়ଶΛϞσϦϯά͢Δ

    View full-size slide

  31. An Ensemble-Based Approach to Click-
    Through Rate Prediction for Promoted Listings
    at Etsy by Kamelia Aryafar et al. ADKDD 2017.
    • "EtsyͷϦεςΟϯά޿ࠂʹ͓͚ΔΞϯαϯϒϧϕʔεCTRਪఆ"

    • ΦϯϥΠϯγϣοϐϯάαΠτEtsyʹ͓͚ΔCTR༧ଌ

    • ϩδεςΟοΫճؼ+FTRL-Proxima

    • ෳ਺ͷํ๏ͰCTRΛਪఆͯ͠ϩδεςΟοΫճؼͰ݁߹

    • աڈσʔλΛεϜʔδϯάͯ͠ਪఆͨ͠CTR(σʔλ͕ଟ͍ͱྑ͘
    ޮ͘)

    • ςΩετσʔλ͸unigram/bigramΛhashing trickͯ͠ϕΫτϧԽ

    • ը૾͸DNNͰຒΊࠐΈ

    View full-size slide

  32. A Practical Framework of Conversion Rate
    Prediction for Online Display
    Advertising by Quan Lu et al. ADKDD 2017.
    • CVR༧ଌʹؔ͢Δ༷ʑͳख๏ͷఏҊ

    • Over prediction΁ͷରԠ

    • delayed feedbackΛߟֶྀͨ͠श

    • ΞτϦϏϡʔγϣϯϞσϧΛߟྀͨ͠ೖࡳֹิਖ਼

    View full-size slide

  33. Ranking and Calibrating Click-Attributed
    Purchases in Performance Display Advertising by
    Sougata Chaudhuri et al. ADKDD 2017.
    • click-attributionͰCV՝ۚϞσϧͷ޿ࠂ഑৴

    • bidͷࡍʹ͸ҎԼͷखॱ͕ඞཁ

    • CV͞Ε΍͍͢޿ࠂͷީิΛྻڍ(ΞτϦϏϡʔγϣϯͷ੍໿
    ͰɺΫϦοΫ཰΋ߴ͍ඞཁ͕͋Δ)

    • CV཰ʹԠͯ͡ೖࡳֹۚΛܾఆ(ਖ਼֬ͳCV཰ͷਪఆ͸ࠔ೉)

    • ϥϯΩϯά+ΩϟϦϒϨʔγϣϯʹΑͬͯ͜ͷ໰୊Λղ͘

    View full-size slide

  34. Deep & Cross Network for Ad Click
    Predictions by Ruoxi Wang et al. ADKDD
    & TargetAd 2017.
    • Google

    • Deep & Cross NetworkϞσ
    ϧͷఏҊ

    • ߴ࣍ͷަޓ࡞༻Λཅʹѻ͏
    ωοτϫʔΫ(Cross)Λ࣋ͭ

    View full-size slide

  35. Field-weighted Factorization Machines for
    Click-Through Rate Prediction in Display
    Advertising by Junwei Pan et al. WWW 2018.
    • Field-aware FM͸ੑೳ͍͍͕ύϥϝʔλ਺ଟա͗ͳͷͰվ
    ળͨ͠

    • ղઆ: https://www.smartbowwow.com/2019/06/field-
    weighted-factorization-machines.html

    View full-size slide

  36. Robust Factorization Machines for User
    Response Prediction by Surabhi Punjabi
    and Priyanka Bhatt. WWW 2018.
    • Webͷಛੑ্ɺϢʔβΛਖ਼֬ʹࣝผͰ͖ͳ͍

    • ผΞϓϦ/σόΠε͔ΒͷΞΫηε

    • CookieແޮԽ

    • ϩόετ࠷దԽͱݺ͹ΕΔख๏ΛFMʹద༻ɺֶशσʔλ
    ͷϊΠζʹڧ͍ϞσϧΛಘΒΕͨ

    • ஶऀʹΑΔղઆ: https://medium.com/@priyankabhatt91/
    robust-factorization-machines-b44e2d906f15

    View full-size slide

  37. A Nonparametric Delayed Feedback Model
    for Conversion Rate Prediction by Yuya
    Yoshikawa and Yusaku Imai. ArXiv 2018.
    • CVRਪఆʹ͓͍ͯ͸ɺϑΟʔυόοΫͷ஗Ԇ͕໰୊ʹͳΔ

    • imp͔ͯ͠Β௕࣌ؒ଴ͨͳ͍ͱCV͢Δ͔Ͳ͏͔Θ͔Βͳ͍

    • ਖ਼ֶ͍͠शͷͨΊʹ͸࣌ؒ஗Εͷ෼෍Λਪఆ͢Δඞཁ͕͋Δ͕ɺ
    ಛఆͷ෼෍ΛԾఆͤͣϊϯύϥϝτϦοΫʹ΍Γ͍ͨ

    • CVRਪఆϞσϧʹϊϯύϥ஗ԆϞσϧΛ૊ΈࠐΜͩख๏ΛఏҊ

    • ೔ຊޠ: https://www.jstage.jst.go.jp/article/pjsai/JSAI2018/0/
    JSAI2018_1N202/_article/-char/ja/

    View full-size slide

  38. ·ͱΊ
    • CTR/CVR༧ଌ͸جຊతʹେن໛ͳςʔϒϧσʔλ

    • εέʔϥϏϦςΟ΍ަޓ࡞༻ͷදݱͳͲ͕ॏཁ

    • Ϟσϧ͸LR͕جຊɺަޓ࡞༻Λѻ͑ΔFM͕҆ఆ

    • ۙ೥͸ը૾΍࣌ܥྻσʔλɺΑΓෳࡶͳަޓ࡞༻Λѻ͏ͨ
    ΊDNNͷಋೖ΋

    View full-size slide