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
スマートフォン向けインターネット広告配信システムの配信最適化
Search
Takashi Nishibayashi
July 11, 2017
Business
3
1.9k
スマートフォン向けインターネット広告配信システムの配信最適化
DATUM STUDIO Conference 2017夏での講演資料です
非エンジニア向けの内容です
Takashi Nishibayashi
July 11, 2017
Tweet
Share
More Decks by Takashi Nishibayashi
See All by Takashi Nishibayashi
論文紹介 Improving Medical Reasoning through Retrieval and Self-Reflection with Retrieval-Augmented Large Language Models
hagino3000
0
720
論文紹介 Audience Size Forecasting Fast and Smart Budget Planning for Media Buyers
hagino3000
0
220
論文紹介 Towards a Fair Marketplace: Counterfactual Evaluation of the trade-off between Relevance, Fairness & Satisfaction in Recommendation Systems
hagino3000
1
610
論文紹介 Budget Management Strategies in Repeated Auctions (公開版)
hagino3000
1
270
論文紹介 A Request-level Guaranteed Delivery Advertising Planning: Forecasting and Allocation
hagino3000
0
100
論文紹介 Online Experimentation with Surrogate Metrics Guidelines and a Case Study
hagino3000
0
260
論文紹介 Bidding Machine: Learning to Bid for Directly Optimizing Profits in Display Advertising
hagino3000
0
170
論文紹介 Balancing Relevance and Discovery to Inspire Customers in the IKEA App
hagino3000
0
710
不確実性と上手く付き合う意思決定の手法
hagino3000
19
15k
Other Decks in Business
See All in Business
見積りと提案の力を競う見積りソン/ an estimation-thon to compete on the quality of estimates and proposals
bpstudy
0
210
SaaSの次なる潮流BPaaS ゼロイチの事業づくりと伴走するプロダクト開発の裏側
kubell_hr
3
1.9k
2ndPASS_リクルートガイド2024
akinodaichi
0
520
CCBJIピッチブック
2024ccbji
0
200
SendGrid Night #10「Email Activityの活用法」
adaisukev
0
190
株式会社Anfini_新卒会社紹介資料
anfini
0
5.7k
メドピアグループ紹介資料
medpeer_recruit
10
120k
Eco-Pork Impact Report 2025.02.09 JP
ecopork
0
460
ユーザーは本当に「AI」を求めている? toCプロダクトにおける生成AI体験づくり事例
inagakikay
1
730
Mercari-Fact-book_en
mercari_inc
1
24k
第3回関東Kaggler会 LT Kaggleはうつ病患者の役に立つ
utm529f
2
210
Space BD - CULTURE BOOK「道」
spacebd
0
1k
Featured
See All Featured
The Illustrated Children's Guide to Kubernetes
chrisshort
48
49k
The Cost Of JavaScript in 2023
addyosmani
47
7.4k
Into the Great Unknown - MozCon
thekraken
35
1.6k
For a Future-Friendly Web
brad_frost
176
9.6k
[RailsConf 2023] Rails as a piece of cake
palkan
53
5.3k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
28
9.3k
VelocityConf: Rendering Performance Case Studies
addyosmani
328
24k
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
40
2k
What’s in a name? Adding method to the madness
productmarketing
PRO
22
3.3k
Making the Leap to Tech Lead
cromwellryan
133
9.1k
Let's Do A Bunch of Simple Stuff to Make Websites Faster
chriscoyier
507
140k
Creating an realtime collaboration tool: Agile Flush - .NET Oxford
marcduiker
27
1.9k
Transcript
εϚʔτϑΥϯ͚ Πϯλʔωοτࠂ৴γεςϜ ʹ͓͚Δ৴࠷దԽ 5BLBTIJ/JTIJCBZBTIJ %"56.456%*0$POGFSFODFՆ
Agenda 1.ࣗݾհ 2.ΞυωοτϫʔΫͱωοτࠂۀքʹ͍ͭͯ 1.ωοτࠂϏδωεͷ֓ཁ 2.ࠂ৴ͰΘΕΔਓೳཁૉٕज़ 3.ฐࣾࣄྫͷհ 1.ΫϦοΫ୯Ձͷௐઅ 2.৴͢Δࠂͷબ
ࣗݾհ ID: hagino3000 Name: ྛ (Takashi Nishibayashi) Job: Software
Engineer ݱࡏZucks AdNetworkʹͯ৴ϩδοΫ ͷ։ൃΛ୲ɻσʔλੳج൫ͷߏங͔Β ػցֶशΛͬͨ༧ଌɺ࠷దԽॲཧ·Ͱɻ
࠷ۙͷ׆ಈ ਓೳֶձࢽ Vol. 32 No. 4 (2017/07) ͷʮࠂͱAI ಛूʯʹຊͷ༰ʹؔ࿈ ͨ͠هࣄΛدߘ͍ͯ͠·͢ɻ
ৄࡉʹڵຯ͕͋Δํ͝Ұಡ͍ͩ͘͞ɻ
Ad Networkͱ ✴ ΠϯλʔωοτͷσΟεϓϨΠࠂྖҬʹ͓͍ͯɺ ෳͷࠂओͱෳͷഔମࣾΛଋͶͯࠂΛ৴͢ ΔΈ ✴ ഔମࣾʹऩӹΛɺࠂओʹίϯόʔδϣϯΛ ͨΒ͢ͷ͕ࣄ ✴
ࠂϦΫΤετຖʹͲͷࠂΛ৴͢Δ͔ϩδοΫ Ͱܾఆ͍ͯ͠Δ
Ad Network ࠂೖߘ ࠂओ ഔମࣾ ࠂഔମ (ϝσΟΞ) ࠂ৴ ࠂඅ
ࠂऩӹ ΦʔσΟΤϯε Click
ωοτࠂ৴ͱ ਓೳཁૉٕज़ ✴ ৴͢Δࠂͷબ ✴ CTRɾCVR༧ଌ ✴ ϢʔβʔτϥοΩϯά ✴ ྫ:
ෳσόΠεΛލ͍ͩߪങߦಈͷ ✴ ࠂޮՌͷਪఆ ✴ ྫ: TV CMͷޮՌਪఆ ✴ ࠂΦʔΫγϣϯʹ͓͚ΔϦΞϧλΠϜೖࡳ
Zucks AdNetworkʹ͓͚Δ ࣄྫհ ✴ લఏ ✴ ৫ͷσʔλ׆༻εςʔδ ✴ Ad NetworkʹٻΊΒΕΔ৴
✴ ࣄྫ1. ΫϦοΫ୯Ձͷ࠷దԽ ✴ ࣄྫ2. ୳ࡧ৴ͷޮԽ
৫ͷσʔλ׆༻εςʔδ σʔλΛཷΊΒΕΔ σʔλ͕ར༻Ͱ͖ͳ͍ॴʹ͍͖ͳΓػցֶशΛͬͨ γεςϜΛσϓϩΠͰ͖ͳ͍ σʔλ͕Ҿ͖ग़ͤΔ ੳ͕Ͱ͖Δ ༧ଌॲཧͷγεςϜԽ ༧ଌ݁ՌΛͬͨऩӹͷ࠷େԽ ݕূͷ
Έ #*πʔϧͷಋೖ "#ςετ ҼՌޮՌਪ ཧ࠷దԽ ػցֶश ੳج൫ͷߏங
ཁһ֬อ ✴ ࠷ॳ͔ΒશͯͷϨΠϠʔͰඞཁͳεΩϧΛ࣋ͭਓࡐ Λἧ͑Δͷ͍͠ ✴ Γͳ͍ॴ͍͍ײ͡ʹิ͍ͬͯ͘ඞཁ͕͋Δ ✴ ֎෦……??
ղ͖͍ͨ ✴ ͍ͭ ✴ ୭ʹ or Ͳͷࠂʹ ✴ ͲͷࠂΛ ✴
(ΫϦοΫ୯Ձ) ͍͘ΒͰ ✴ දࣔ͢Δͷ͔
ఆࣜԽ ✴ ඪ ✴ ഔମࣾऩӹͷ࠷େԽ ✴ ੍݅ ✴ ࠂओͷඪCPA (ίϯόʔδϣϯ֫ಘ͋ͨΓͷίετ)
✴ ࠂओͷ༧ࢉ ✴ ࠂදࣔճ ͨͩ͠ ΫϦοΫɾίϯόʔδϣϯ ৴͠ͳ͍ͱΘ͔Βͳ͍
Ұͭͷ࠷దԽͱͯ͠ղ͚Εྑ͍ ͷͰ͕͢ɺ͍͠ͷͰෳͷʹ ͚ͯ։ൃͯ͠·͢
ΫϦοΫ୯Ձͷௐ ✴ CPA (ίϯόʔδϣϯ͋ͨΓͷ֫ಘίετ) Λࠂओ ͷཁʹ߹ΘͤΔͷ͕త ✴ ͋ΔࠂΩϟϯϖʔϯΛ৴͢Δͱͯ͠ ✴ ίϯόʔδϣϯ͕औΕΔͷ୯Ձ্͍͛ͨ
՝ ✴ ྫ ✴ ίϯόʔδϣϯ100% ✴ ΫϦοΫ୯Ձ100ԁͳΒCPA100ԁͱͳΔ ✴ ͳΔ͘৴ͷॳظஈ֊ʹ͓͍ͯίϯόʔδϣϯ Λਪఆ͍ͨ͠
✴ ͔͠͠ɺ৴ॳظΫϦοΫͷαϯϓϧαΠζ͕খ ͘͞౷ܭతʹྑ͍ͱѱ͍ͱݴ͑ͳ͍
CVRਪఆ ✴ ίϯόʔδϣϯͷࣅͨಉ࢜Ͱ͋ΕɺCVRۙ͘ ͳΔͣɻ͜ΕΛࣄલͱͯ͑͠ͳ͍͔ ✴ ࣅͨಉ࢜ͷू߹ΫϥελϦϯάͰٻΊΔ ✴ ࣄલ֬Λಋೖ͠ɺϕΠζͷఆཧʹΑΓΫϦοΫ n ͷ
͏ͪ k ݸͷίϯόʔδϣϯΛ؍ଌͨ͠ޙͷ CVR ͷࣄޙ ֬Λߟ͑ΔɻCVRͷࣄલΛϕʔλBeta(α, β) ͱ͢ΔͱɺCVRͷࣄޙϕʔλʹͳΔɻ
݁Ռݕূ1 ✴ ༧ଌਫ਼ΦϑϥΠϯ࣮ݧͰݕূͰ͖Δ ✴ RMSE, Accuracy, Precision, F-value …. ✴
ϏδωεαΠυ͕Γ͍ͨͷɺ༧ଌ͕ͨΔࣄʹΑ ΔܦӦࢦඪͷӨڹ (ྫ: ച্) ༧ଌਫ਼͕YY্͕Γ·ͨ͠ ച্Ͳ͏ͳΔͷʜʜ
݁Ռݕূ2 ✴ ࣮ࡍʹCPA͕ඪCPAʹۙ͘ͳΔͷ͔ɺຊ൪ʹϦϦʔ εͯ͠ݕূ ✴ log(࣮CPA/ඪCPA) ΛطଘϩδοΫద༻Ωϟϯ ϖʔϯͱൺֱɻରͰݟΔͷɺ2ഒʹͳΔͷͱ ʹͳΔͷΛಉ͡ΠϯύΫτͱͯ͠ଊ͑ΔͨΊɻ ✴
log(࣮CPA/ඪCPA) ͷʹ͍ͭͯϊϯύϥϝτ ϦοΫݕఆͰ͕ࠩ͋Δࣄͷ֬ೝͱ4ҐͷࠩΛΈΔ
ެ։൛ʹ͖ͭআ ݁Ռ
৴͢Δࠂͷબ ✴ ഔମऀऩӹͷߴ͍ࠂΛଟ͘৴͍ͨ͠ ✴ ΫϦοΫ͕ଟ͘ίϯόʔδϣϯऔΕΔ ✴ ݁ՌతʹΫϦοΫ୯Ձ্͛ΒΕΔ ✴ ޮՌ͕ྑ͍͔ѱ͍͔৴͠ͳ͍ͱΘ͔Βͳ͍ ✴
ࠂͱࠂͷΈ߹Θͤແʹ͋ΔͷͰૣ͘ྑ ͍Έ߹ΘͤΛҾ͖͍ͯͨ ✴ ࣝͷ׆༻ͱ୳ࡧͷδϨϯϚ
୳ࡧͱ׆༻ ✴ ׆༻ ✴ ʹͱͬͯऩӹ͕ߴ͍ͱΘ͔͍ͬͯΔࠂΛ৴ ✴ ࠷ߴ͍ͷΈΛ৴ͨ͠Βྑ͍༁Ͱͳ͍ ✴ طଘͷόϯσΟοτΞϧΰϦζϜΛͦͷ··͍ ʹ͍͘
✴ ୳ࡧ ✴ ʹͱͬͯऩӹ͕ະͷࠂΛ৴͢Δ
ଟόϯσΟοτʹΑΔ Ξϓϩʔν ✴ εϩοτϚγϯͷϝλϑΝʔ ✴ εϩοτϚγϯ͕ෳ͋ͬͨ࣌ʹͲΕΛԿճҾ͘ ͖͔ ✴ ϝϦοτ ✴
ڭࢣσʔλ͕ແ͍ॴ͔ΒελʔτͰ͖Δ ✴ ৽͍͠ࠂΩϟϯϖʔϯ͕࣍ʑͱೖߘ͞ΕΔઃఆ
୳ࡧͷޮԽ ✴ ͋Δʹ͓͚Δɺଞͷࠂͱͷൺֱ ✴ ଞͷࠂͱൺֱͯ͠ऩӹ͕ѱ͍ͱΘ͔ͬͨ࣌Ͱ୳ ࡧΛΊΕྑ͍ ✴ ऩӹੑ(eCPM)ͷ্քΛ͏ ✴ ֬తʹߴͯ͘͜Ε͙Β͍ͩΖ͏ɺͱ͍͏
✴ ৴Λଓ͚ΔࣄͰԼ͕͍ͬͯ͘
ݕূ ✴ ͷऩӹ͕Ͳ͏มԽ͔ͨ͠Λݟ͍ͨ ✴ ͔͠͠ɺऩӹ࣌ؒมԽͷӨڹΛڧ͘ड͚Δ ✴ ظʹ༧ࢉফԽ͕͋ΔͨΊɺඞͣ৳ͼΔ ✴ ୯७ʹϩδοΫมߋલޙͰൺֱͰ͖ͳ͍
ϥϯμϜԽൺֱࢼݧʹΑΔݕূ ϩδοΫͷมߋʹΑΔհೖ σʔλαϯϓϧɺԣ࣠࣌ؒ ࠂΛ܈ʹ͚ͯσʔλΛऔΔ
݁Ռ ެ։൛ʹ͖ͭআ
͋Γ͕ͱ͏͍͟͝·ͨ͠