Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
オンライン広告関連の論文を50本くらい雑に紹介する AdKDD編 / adkdd-all
Search
todesking
December 13, 2019
4
2.2k
オンライン広告関連の論文を50本くらい雑に紹介する AdKDD編 / adkdd-all
todesking
December 13, 2019
Tweet
Share
More Decks by todesking
See All by todesking
自作言語進捗 2020 Mar / ojaml-2020-mar
todesking
0
360
バンディット問題の理論とアルゴリズム 第8章 / bandit-8
todesking
0
89
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
3
1.6k
バンディット問題の理論とアルゴリズム 第二章 / bandit2
todesking
0
120
自作言語進捗 2019 May / ojaml-2019-may
todesking
0
900
自作言語進捗 2019 Mar
todesking
0
450
ベイズ統計モデリング 10 // Doing Bayesian Data Analysis Chapter 10
todesking
0
99
実行時におけるJVMバイトコード最適化手法
todesking
16
12k
Featured
See All Featured
Speed Design
sergeychernyshev
24
570
What's in a price? How to price your products and services
michaelherold
243
12k
Ruby is Unlike a Banana
tanoku
96
11k
4 Signs Your Business is Dying
shpigford
180
21k
How GitHub (no longer) Works
holman
311
140k
The Straight Up "How To Draw Better" Workshop
denniskardys
232
140k
XXLCSS - How to scale CSS and keep your sanity
sugarenia
246
1.3M
It's Worth the Effort
3n
183
27k
Designing on Purpose - Digital PM Summit 2013
jponch
115
6.9k
The Pragmatic Product Professional
lauravandoore
31
6.3k
The Power of CSS Pseudo Elements
geoffreycrofte
72
5.3k
Java REST API Framework Comparison - PWX 2021
mraible
PRO
28
7.9k
Transcript
ΦϯϥΠϯࠂؔ࿈ͷจΛ 50ຊ͘Β͍ࡶʹհ͢Δ AdKDDฤ @todesking
ࣗݾհ • @todesking • DSP/ΞυωοτϫʔΫͷձࣾ ͰػցֶशΛ͍ͬͯΔ • ࠒͷٕज़ௐ͕ࠪΓͳ͍ͳ ͱࢥͬͨͷͰAdKDDͷจΛ ோΊͯΈ·ͨ͠
AdKDD/TargetADͱ • KDD: Knowledge Discovery and Data Mining • σʔλϚΠχϯάػցֶशʹؔ͢ΔԠ༻తͳΛѻ͏ࠃࡍ
ձٞ • AdKDD: ΦϯϥΠϯࠂؔ࿈ٕज़Λѻ͏Պձ • TargetAd: WSDMͱ͍͏ࠃࡍձٞʹ͓͍ͯΦϯϥΠϯࠂؔ࿈ٕज़ Λѻ͏Պձ • ࠷ۙAdKDDʹٵऩ͞Εͨ? • ࠾จຖ10ຊऑ • KDDຊମʹࠂܥจ͕͋Δ(Έ͚Ṗ)
ҙ • ཁ΄΅AbstractோΊ͚ͨͩɺਖ਼֬͞อূ͠·ͤΜ • ؒҧ͍ͷࢦఠܴ • ۙͷจެࣜαΠτ͔ΒPDFͷϦϯΫ͕͋Δ • Keynote/Invited talkলུ
AdKDD 2014 https://dblp.org/db/conf/kdd/adkdd2014
A Dynamic Pricing Model for Unifying Programmatic Guarantee and Real-Time
Bidding in Display Advertising • publisher͕impΛചΔʹRTBͱguranteed contractͱ͍ ͏ೋͭͷํ๏͕͋Δ • publisherͷऩӹΛදݱ͢ΔཧϞσϧΛ࡞Γɺೋͭͷํ ๏ͷ༧ࢉΛ࠷దԽ͢Δ
Multi-Touch Attribution Based Budget Allocation in Online Advertising • Ωϟϯϖʔϯશମͷ༧ࢉ͕ܾ·͓ͬͯΓɺԼҐͷΩϟϯ
ϖʔϯʹ͍ͨ͠ • ඪͷߩݙʹԠ͍ͯͨ͡͠ˠlast-touch/multi- touch attributionͷใΛݩʹׂͯ͠ΓͯΔํ๏Λ࣮ ͨ͠
Pleasing the advertising oracle: Probabilistic prediction from sampled, aggregated ground
truth • ࠂΩϟϯϖʔϯͷηάϝϯτਫ਼Λୈࡾऀ͕ධՁ͢Δ έʔε • ධՁͱͯ͠ɺࠂΛݟͨϢʔβͷσϞάϥଐੑͷׂ߹͕ ϑΟʔυόοΫ͞ΕΔ • ݸผϢʔβͷଐੑʹ͍ͭͯͷใಘΒΕͳ͍ • ͜ͷใΛͬͯηάϝϯτͷਫ਼Λվળ͢Δ
The PlaceIQ Analytic Platform: Location Oriented Approaches to Mobile Audiences
• PlaceIQͱ͍͏Ґஔϕʔεͷࠂੳπʔϧͷհ
Practical Lessons from Predicting Clicks on Ads at Facebook •
ஶ໊ͳΦϯϥΠϯࠂจ • FBͷCTR༧ଌϞσϧͷ࣮ʹ͍ͭͯ
iPinYou Global RTB Bidding Algorithm Competition Dataset • iPinYouͱ͍͏DSP͕ओ࠵ͨ͠RTBΞϧΰϦζϜͷίϯς ετͰ༻͞Εͨσʔληοτʹ͍ͭͯ
• ͜ͷσʔληοτRTBܥϩδοΫͷ࣮ݧʹΑ͘ΘΕͯ ͍Δ
Can Television Advertising Impact Be Measured on the Web? Web
Spike Response as a Possible Conversion Tracking System for Television. • ςϨϏࢹௌऀಉ࣌ʹΠϯλʔωοτΔͷͰɺςϨϏ CMͷࢹௌ͕ेඵޙͷωοτ׆ಈʹӨڹΛ༩͑Δ • ͜ͷΑ͏ͳӨڹΛܭଌ͢ΔͨΊͷwebϕʔεγεςϜΛ։ ൃͨ͠
TargetAd 2016 https://sites.google.com/site/targetad2016/program
TargetAd 2016 https://sites.google.com/site/targetad2016/program
TargetAd 2016 https://sites.google.com/site/targetad2016/program
Distributed Representations of Web Browsing Sequences for Ad Targeting •
ϢʔβͷϒϥδϯάཤྺΛಛྔԽ͍ͨ͠ • ϢʔβΛparagraphɺURLΛwordʹݟཱͯͯParagraph Vector(W2VͷੜɺύϥάϥϑͷembeddingΛಘΒΕΔ) Λߏங͢Δͱ͍͏ͷΛલճͬͨ • Backward PV-DMͱ͍͏վྑ൛ͷఏҊ
Sponsored Video Advertising: Filling the Void between Online and TV
Advertising • Ϣʔβ͕ಈըΛࢹௌ͍ͯ͠Δͱ͖ʹࠂΛݟ͍ͤͨɺݟͤ ·͘Γ͍ͨͱ͍͏ڧ͍ؾ͕࣋ͪ͋Δ • ಈըʹؔ࿈͢Δࠂͷ՝ۚମܥʹ͍ͭͯͷఏҊ
On Spatial-Aware Viral Marketing for Location-based Advertisements, • ৽͍͠ళΛ։͖͍ͨ߹ɺళͷॴɺޱίϛͷޮՌɺ ཧࠂͷॴͳͲͷཁૉ͕ച্͛ʹӨڹΛ༩͑Δɻ
• ͦͷΑ͏ͳ੍ͷԼͰ࠷దͳީิΛબͿํ๏ΛఏҊ
Keynote: Come back soon! Estimating return times for users •
ߦಈཤྺΛݩʹɺϢʔβ͕࣍ʹ͜ͷαʔϏεΛ͏ͷ ͍͔ͭਪఆ͍ͨ͠ • ੜଘੳɺաఔɺLSTM
Building a Bi-directed Recommendation System for Mobile Users and App-install
Ad Campaigns • ϞόΠϧΞϓϦʹؔ͢ΔํϨίϝϯσʔγϣϯγες Ϝ • ΞϓϦͷࠂΩϟϯϖʔϯʹରͯ͠ɺޮՌతͳϢʔβηά ϝϯτΛਪન • Ϣʔβʹରͯ͠ɺڵຯΛ࣋ͪͦ͏ͳΞϓϦΛਪન
Finding Needle in a Million Metrics: Anomaly Detection in a
Large-scale Computational Advertising Platform • ࠂϓϥοτϑΥʔϜʹ͓͚Δҟৗݕ • Ξυϑϥυͷจ຺Ͱͳ͘ɺγεςϜͷҟৗΛݕग़͢Δ ͷ͕త • ҆ఆͨ͠ύϑΥʔϚϯεΛग़͍ͯ͠ΔΩϟϯϖʔϯΛ؍ଌ • ύϑΥʔϚϯεʹҟৗ͕͋ͬͨΒɺγεςϜଆʹݪҼ͕ ͋ΔՄೳੑ͕ߴ͍
Preserving Privacy in Geo- Targeted Advertising • ҬͱऩೖΛͬͨλʔήςΟϯάࠂΛߟ͑Δ • ߈ܸऀτϐοΫϞσϦϯάػցֶशΛ͏͜ͱͰɺ͋
Δࠂ͕දࣔ͞ΕͨϢʔβͷҬऩೖΛਪఆͰ͖Δ • ৴͞ΕΔࠂʹϊΠζΛࠞͥΔ͜ͱͰ͜ͷΑ͏ͳਪఆΛ ͛Δ
Invited talk: User Recommendation in Instagram • Instagramʹ͓͚Δ͓͢͢ΊϢʔβͷΈ • ιʔγϟϧάϥϑͷߏڵຯͷ͋ΔτϐοΫΛͬͨਪ
ન
Demographic Prediction of Web Requests from Labeled Aggregate Data •
Ϣʔβू߹ʹରͯ͠ɺσϞάϥଐੑͷׂ߹ΛΔ͜ͱ͕Ͱ ͖Δͱ͢Δ • ͜ͷใΛͬͯϢʔβ୯ҐͰͷଐੑਪఆΛ͢Δ • Williams+, "Pleasing the advertising oracle: Probabilistic prediction from sampled, aggregated ground truth", AdKDD2014 ͷൃల
Toward Personalized Product Search for eCommerce Sites: A Case Study
in Yahoo! Taiwan • Y!ͷeίϚʔεαΠτʹ͓͚Δݕࡧ • ଟ͘ͷݕࡧΫΤϦҰൠతͩͬͨΓᐆດͰɺϢʔβେ ͳ݁Ռͷத͔ΒతͷͷΛ୳͞ͶͳΒͳ͍ • ςΩετใ͚ͩͰͷϚονϯάͰෆेͳͷͰɺϢʔ βͷڵຯʹԠͯ͡ύʔιφϥΠζͨ͠ϥϯΩϯάΛߦ͏
AdKDD & TargetAd 2017
AdKDD & TargetAd 2017 https://adkdd-targetad.wixsite.com/adkddtargetad2017/accepted-papers
AdKDD & TargetAd 2017 https://adkdd-targetad.wixsite.com/adkddtargetad2017/accepted-papers
AdKDD & TargetAd 2017 https://adkdd-targetad.wixsite.com/adkddtargetad2017/accepted-papers
Attribution Modeling Increases Efficiency of Bidding in Display Advertising •
Criteo • ΞτϦϏϡʔγϣϯϞσϧΛͬͨbidઓུ • 2nd price auctionʹ͓͚Δ࠷దઓུͦͷϦΫΤετͷՁ Λೖࡳֹͱ͢Δ͜ͱ͕ͩɺਖ਼֬ͳՁΛٻΊΔͨΊʹ ΞτϦϏϡʔγϣϯϞσϧΛߟྀ͢Δඞཁ͕͋Δ
Blacklisting the Blacklist in Online Advertising • Dstillery • RTBʹ͓͍ͯpublisherଆ͕ϒϥοΫϦετΛ͍࣋ͬͯͯ
ಛఆͷࠂΛ͘͜͡ͱ͕͋Δ • ϒϥοΫϦετͷ༰ඇެ։ • ແବͳbidΛආ͚ΔͨΊϒϥοΫϦετͷ༰Λਪఆ͢Δ ػೳΛ࡞ͬͨ • γεςϜෛՙ͕ݮΓɺউ্͕ͨ͠
Anti-Ad Blocking Strategy: Measuring its True Impact • Adobe Research
• AdBlockΛݕग़ͯ͠ϢʔβʹϝοηʔδΛग़͢ઓུ͕͋Δ (՝ۚίʔεͷ༠ಋͳͲ) • ͜ͷઓུͲͷఔ༗ޮͳͷ͔ௐΔͨΊͷ౷ܭతख๏ ΛఏҊ • ͳΜΒ͔ͷࣄʹΑΓɺී௨ʹA/Bςετ͢ΔΘ͚ʹ͍ ͔ͳ͍Β͍͠
MM2RTB: Bringing Multimedia Metrics to Real-Time Bidding • ࠂͷ࣭ΛධՁ͢ΔͨΊͷmultimedia metricsͱ͍͏ͷ͕
͋Δ • contextual relevance, visual saliency, ad memorability • ϖʔδʹෳͷࠂεϩοτ͕͋Δͱ͖ɺmultimedia metricsΛߟྀͯ͠ࠂͷΈ߹ΘͤΛબ͢Δख๏ͷఏ Ҋ
Data-Driven Reserve Prices for Social Advertising Auctions at LinkedIn •
ηΧϯυϓϥΠεΦʔΫγϣϯʹ͓͚Δ࠷దͳϦβʔϒϓ ϥΠεΛٻΊΔํ๏Λ2छྨ(Ϣʔβ/ηάϝϯτϨϕϧ)ఏ Ҋ • ࣮γεςϜʹ࣮ͯ͠ޮՌΛݕূͨ͠
Optimal Reserve Price for Online Ads Trading Based on Inventory
Identification • Alibaba+Yahoo • ͜Εreserve priceͷܾఆઓུʹؔ͢Δ
Ranking and Calibrating Click-Attributed Purchases in Performance Display Advertising •
click-attributionͰCV՝ۚϞσϧͷࠂ৴ • bidͷࡍʹҎԼͷखॱ͕ඞཁ • CV͞Ε͍͢ࠂͷީิΛྻڍ(ΞτϦϏϡʔγϣϯͷ ੍ͰɺΫϦοΫߴ͍ඞཁ͕͋Δ) • CVʹԠͯ͡ೖࡳֹۚΛܾఆ(ਖ਼֬ͳCVͷਪఆࠔ ) • ϥϯΩϯά+ΩϟϦϒϨʔγϣϯʹΑͬͯ͜ͷΛղ͘
Cost-sensitive Learning for Utility Optimization in Online Advertising Auctions •
CTR/CVR༧ଌʹɺhighly non-uniform misprediction cost͕ଘࡏ͢Δ • UtilityͷΑ͏ͳࢦඪΛͬͯੑೳධՁ͢Δख๏͕ఏҊ͞Ε ͍ͯΔ • ֶश࣌ʹUtilityΛߟྀͯ͠࠷దԽ͍ͨ͠ • log lossʹॏΈΛ͚ͭΔ͜ͱͰੑೳվળ͢Δख๏ͷఏҊ
A Practical Framework of Conversion Rate Prediction for Online Display
Advertising • CVR༧ଌʹؔ͢Δ༷ʑͳख๏ͷఏҊ • Over predictionͷରԠ • delayed feedbackΛߟֶྀͨ͠श • ΞτϦϏϡʔγϣϯϞσϧΛߟྀͨ͠ೖࡳֹิਖ਼
An Ensemble-based Approach to Click- Through Rate Prediction for Promoted
Listings at Etsy • Etsyͷlisting adͰΘΕ͍ͯΔCTRਪఆγεςϜͷհ
Profit Maximization for Online Advertising Demand-Side Platform • CPC/CPAϞσϧʹ͓͍ͯɺDSPͷརӹΛ࠷େԽ͍ͨ͠ •
࠷దͳೖࡳઓུ(ͲͷࠂΛ͍͘ΒͰೖࡳ͢Δ͔)ʹ͍ͭͯ ߟ • Ωϟϯϖʔϯ༧ࢉͱϖʔγϯάɺλʔήςΟϯάɺimpڙ څྔͷ੍͕͋Δ • ࣮༻తʹղ͚ΔϞσϧΛఏҊ
Deep & Cross Network for Ad Click Predictions • Google
• Deep & Cross networkͱ͍͏DLͷϞσϧΛఏҊ • ަޓ࡞༻Λཅʹѻ͑Δ • CTRਪఆʹར༻ͯ͠ߴੑೳͩͬͨ
AdKDD & TargetAd 2018
AdKDD & TargetAd 2018 https://adkdd-targetad.wixsite.com/2018/accepted-papers
AdKDD & TargetAd 2018 https://adkdd-targetad.wixsite.com/2018/accepted-papers
A Large Scale Benchmark for Uplift Modeling • Criteo •
Uplift modelingͷݕূʹ͑ΔେنσʔλΛެ։ͨ͠ • https://ailab.criteo.com/criteo-uplift-prediction-dataset/
Deep Density Networks and Uncertainty in Recommender Systems • Taboola(ࠂϓϥοτϑΥʔϜΛͬͯΔձࣾ)
• Deep Density Networksͱ͍͏ϞσϧͷఏҊ • content based+ڠௐϑΟϧλϦϯάͷϋΠϒϦου • ༧ଌͷෆ࣮֬ੑΛදݱͰ͖Δ • ε-greedyͱUCBΛΈ߹ΘͤͨΑ͏ͳϩδοΫͰ exploration/exploitation͢Δ
Deep Neural Net with Attention for Multi-channel Multi-touch Attribution •
DNNΛͬͨMulti-touch attributionϞσϧͷఏҊ
Deep Policy Optimization for E-commerce Sponsored Search Ranking Strategy •
2018͔ΒDNNΛͬͨൃදཱ͕ͪ·͢Ͷ • εϙϯαʔυαʔνͷϥϯΩϯάΛਂڧԽֶशͰ ղ͘
Designing Experiments to Measure Incrementality on Facebook • Facebookʹ͓͚Δincrementalityܭଌख๏ͷղઆ •
upliftͱincrementalityͷҧ͍͕Ṗ • ୯ͳΔA/BςετΑΓॊೈͳ࣮ݧ͕Մೳ
Dynamic Hierarchical Empirical Bayes: A Predictive Model Applied to Online
Advertising • Adobe • εϙϯαʔυαʔνʹ͓͚ΔCTR/CVRਪఆϞσϧ • ֊ϕΠζʹ͓͍ͯɺσʔλʹԠͯ͡ಈతʹ֊ߏΛܾ ఆ͢ΔΑ͏ͳϞσϧͷఏҊ
Forecasting Granular Audience Size for Online Advertising • ಛఆͷηάϝϯτʹࠂ৴͢Δͱ͖ɺࣄલʹΦʔσΟΤ ϯεͷαΠζ͕Γ͍ͨ
• Frequent Itemset Miningͱ͍͏ख๏ΛݩʹɺͦͷΑ͏ͳਪ ఆΛ͢Δख๏ΛఏҊ
Mini-Batch AUC Optimization • ֶश࣌ʹAUCΛ࠷దԽ͍͕ͨ͠ɺେنσʔληο τʹ͓͍͔ͯͬͨ͠ • ϛχόονΛ͏͜ͱͰߴͳֶशΛՄೳͱ͢Δख๏Λ ఏҊ
Optimal Bidding, Allocation and Budget Spending for a Demand Side
Platform Under Many Auction Types • ৽͍͠ೖࡳઓུϩδοΫͷఏҊ • 1st/2nd price auctionʹରԠɺ༧ࢉ੍ΫϥΠΞϯτ͝ ͱͷرΛߟྀͰ͖Δ
AdKDD 2019
AdKDD 2019 https://www.adkdd.org/2019-papers-and-talks
AdKDD 2019 https://www.adkdd.org/2019-papers-and-talks
AdKDD 2019 https://www.adkdd.org/2019-papers-and-talks
Time-Aware Prospective Modeling of Users for Online Display Advertising •
Yahoo • ࣌ܥྻΛߟྀͨ͠ϢʔβߦಈϞσϦϯάʹΑΓɺϢʔβͷ ҙਤΛߟྀͨ͠ਪఆΛߦ͏DNNϞσϧͷఏҊ
Graphing Crumbling Cookies • ΫϩεσόΠετϥοΩϯάΛ͍͕ͨ͠ɺۙcookieͷ ༻੍ݶ͕ݫ͘͠ͳ͍ͬͯΔ(ITP) • Browser fingerprintingͳͲͷΘΔ͍ٕज़ΛΘͣ͜ͷঢ়گ ʹରԠ͢ΔͨΊͷख๏ΛఏҊ
Causally Driven Incremental Multi Touch Attribution Using a Recurrent Neural
Network • JD.com(ژ౦): தࠃͷڊେECاۀ • multi-touch attributionΛRNNͰϞσϦϯά͢Δख๏Λఏ Ҋ • ͜ͷϞσϧ࣮ࡍΘΕ͍ͯΔΒ͍͠
Combinatorial Keyword Recommendations for Sponsored Search with Deep Reinforcement Learning
• Baidu • ਂڧԽֶशͰϨίϝϯμ࡞Δͭ
Feasible Bidding Strategies through Pure Exploration Bandits • ೖࡳઓུͷީิ͕େྔʹ͋Γɺ͍͍ͭΛબͼ͍ͨ •
ύϥϝλͷΈ߹ΘͤͰ͔ͳΓީิ͕ଟ͍(~100)ͨΊɺී ௨ʹόϯσΟοτΞϧΰϦζϜΛ͏ͷඇޮ • ʮͦͦ͜͜Αͦ͞͏ʯͳީิΛߴʹൃݟ͢ΔͨΊͷख๏ ΛఏҊ
In-app Purchase Prediction Using Bayesian Personalized DwellDay Ranking • ژେ+CyberAgent
• ΞϓϦ՝ۚͦ͠͏ͳϢʔβΛਪఆ͢ΔλεΫ • Bayesian personalized rankingΛݩʹͨ͠ख๏ΛఏҊ
Learning from Multi-User Activity Trails for B2B Ad Targeting •
Yahoo • B2Bʹ͓͚Δͷߪೖϓϩηεɺෳͷਓ͕ؒؔΘͬ ͨෳࡶͳʹͳΔ • ͜ͷΑ͏ͳϓϩηεʹରԠͨ͠ࠂͷ৴ઓུ͕ඞཁ • ಉ͡৫ʹ͍ΔΦʔσΟΤϯεΛਪఆɺͦΕΒͷߦಈΛ૯ ߹ͯ͠CV༧ଌ͢ΔϞσϧΛఏҊ
Modeling Advertiser Bidding Behaviors in Google Sponsored Search with a
Mirror Attention Mechanism • SSPଆ͔ΒೖࡳऀͷߦಈΛϞσϦϯά͢Δ • ৽ػೳΛϦϦʔεͨ͠ࡍɺͦΕ͕ظతʹࠂओʹͲͷΑ ͏ͳӨڹΛٴ΅͔͢Γ͍ͨ…… ͱ͍͏ͷ͕Ϟνϕʔ γϣϯΒ͍͠
Optimal bidding: a dual approach • ৽͍͠ೖࡳઓུͷఏҊ
·ͱΊ • ࠂܥจ͕ͨ͘͞Μ͋Δ͜ͱ͕Θ͔Γ·ͨ͠ • ͍͔͕Ͱ͔ͨ͠? • དྷ಄͘Β͍ʹࠂܥจಡॻձΛΓ͍ͨͷͰɺڵຯ͋ Δਓ͓͕͚͍ͩ͘͞