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
Review: "Recommending Investors for Crowdfundin...
Search
Sponsored
·
Your Podcast. Everywhere. Effortlessly.
Share. Educate. Inspire. Entertain. You do you. We'll handle the rest.
→
yag_ays
July 09, 2014
Research
1.2k
1
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Review: "Recommending Investors for Crowdfunding Projects"
http://yagays.github.io/blog/2014/07/09/www2014review-kickstarter/
yag_ays
July 09, 2014
More Decks by yag_ays
See All by yag_ays
対話型AIの構築における工夫とデータセットの重要性 - 素早くデータを構築し検証するためには
yag_ays
3
7.1k
目と耳を持った自然言語処理 - スタートアップにおける価値創出のために
yag_ays
1
3.9k
時間情報表現抽出とルールベース解析器のこれから / Temporal Expression Analysis in Japanese and Future of Rule-based Approach
yag_ays
1
2.3k
Pythonで始める ドキュメント・インテリジェンス入門 / Introduction to Document Intelligence with Python
yag_ays
9
9.2k
"医者の言葉、患者の言葉、エンジニアの言葉" / MNTSQ Ubie Vertical ai
yag_ays
3
14k
LT at nlp_career
yag_ays
0
350
Other Decks in Research
See All in Research
Any-Optical-Model: A Universal Foundation Model for Optical Remote Sensing
satai
3
830
Ghost in the 7‑Zip: The Shadow of Residential Proxies Creeping into Your Life
nttcom
0
1.2k
衛星×エッジAI勉強会 衛星上におけるAI処理制約とそ取組について
satai
4
560
適応的スパムフィルタのための軽量な類似メッセージカウンタ / jsai2026-adaptive-spam-filter
monochromegane
0
3.7k
Apache Gravitinoで実現する Icebergカタログ統合とアクセスの一元化
matsumooon
0
290
Claude Code × autoresearch 実践
mathbullet
0
160
NLP colloquium: AI Safety Survey
kanekomasahiro
0
730
Model Discovery and Graph Simulation: A Lightweight Gateway to Chaos Engineering
anatolykr
0
200
SoftMatcha 2: 1兆語規模コーパスの超高速かつ柔らかい検索
e869120_sub
6
3.5k
「車1割削減、渋滞半減、公共交通2倍」を 熊本から岡山へ@RACDA設立30周年記念都市交通フォーラム2026
trafficbrain
1
1.2k
第66回コンピュータビジョン勉強会@関東 Epona: Autoregressive Diffusion World Model for Autonomous Driving
kentosasaki
0
630
Unified Audio Source Separation (Defense Slides)
kohei_1979
1
610
Featured
See All Featured
Accessibility Awareness
sabderemane
1
140
Unlocking the hidden potential of vector embeddings in international SEO
frankvandijk
0
840
How to Grow Your eCommerce with AI & Automation
katarinadahlin
PRO
1
210
How to train your dragon (web standard)
notwaldorf
97
6.7k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
49
3.5k
Imperfection Machines: The Place of Print at Facebook
scottboms
270
14k
SEOcharity - Dark patterns in SEO and UX: How to avoid them and build a more ethical web
sarafernandez
0
200
JAMstack: Web Apps at Ludicrous Speed - All Things Open 2022
reverentgeek
1
470
Improving Core Web Vitals using Speculation Rules API
sergeychernyshev
21
1.5k
Evolving SEO for Evolving Search Engines
ryanjones
0
220
Building a Modern Day E-commerce SEO Strategy
aleyda
45
9.1k
So, you think you're a good person
axbom
PRO
2
2.1k
Transcript
Recommending Investors for Crowdfunding Projects WWW 2014 Jisun An, Daniele
Quercia, Jon Crowcroft จհ @yag_ays 1
ࠓճհ͢Δจͷ֓ཁ • “Recommending Investors for Crowdfunding Project” [Jisun+ 2014] •
WWW 2014 (Seoul, KOREA) • Jisun AnͷYahoo Labs in Barcelona Πϯλʔϯγοϓͷࣄ ! • ࠷ऴతͳඪɿKickstarterͷϑΝϯμʔͱग़ࢿऀͷϚονϯά • KickstarterͷϓϩδΣΫτग़ࢿऀͷੳϝΠϯͳͱ͜Ζ͕͋Δ 2
• ΫϥυϑΝϯσΟϯά • 2012ʹूΊͨࢿ૯ֹ$320 million • c.f. ࢿՈ/ϕϯνϟʔΩϟϐλϧʹΑΔग़ࢿ ! •
ϑΝϯμʔඪֹۚΛઃఆͯ͠ࢿΛืΔ • ࢿՈࢿֹۚʹԠͯ͡ใु͕Β͑Δ • e.g. $100ࢿͰ1ݸϓϨθϯτɼ$300ࢿͰ5ݸϓϨθϯτ Kickstarterͱ https://www.kickstarter.com/help/style_guide 3
Kickstarterޭࣄྫ • Oculus Rift • 9,522 / $ 2,437,429 •
Memoto (Narrative Clip) • 2,871 / $ 550,189 • Little Witch Academia 2 • 7,938 / $ 625,518 https://www.kickstarter.com/projects/1523379957/oculus-rift-step-into-the-game https://www.kickstarter.com/projects/martinkallstrom/memoto-lifelogging-camera https://www.kickstarter.com/projects/1311401276/little-witch-academia-2 4
Kickstarterͷಛघੑɿࢿʹࣦഊͨ͘͠ͳ͍͚Ͳ… • All or Nothing • ඪֹۚʹୡ͠ͳ͚ΕϓϩδΣΫτࣦഊɼࢿۚશֹฦ٫ • ϓϩδΣΫτͷޭ/ࣦഊʹؔΘΒͣɼࢿऀଛΛ͠ͳ͍ !
• ιʔγϟϧͳଆ໘ • ॳظࢿʹ༑ୡ͕ଟ͍ʢ20-40%ͱ͍͏ࢉग़ʣ • ेͳࢿՈΛूΊΒΕͳ͍ͱϓϩδΣΫτ͕ࣦഊ͍͢͠ 5
จͷྲྀΕ • Kickstarterʹ͓͚ΔࢿՈͷڍಈʹ͍ͭͯԾઆΛཱͯΔ • KickstarterTwitterͷใ͔ΒԾઆΛݕূ͢Δ • ࣗಈతʹϑΝϯμʔͱࢿՈͷϚονϯάΛߦ͏ϞσϧΛཱͯΔ 6
Kickstarterͷσʔλऩू/ղੳ Dataset and Pledging Behavior 7
σʔληοτ • Kickstarter͔ΒΫϩʔϧ • 20137݄͔Β10݄ʹొ͞Εͨͷ • USA෦ͷϑΝϯυͷΈ • ߹ܭ 1,149ϓϩδΣΫτ/
78,460ग़ࢿऀ • Twitter͔ΒΫϩʔϧ • ϓϩδΣΫτʹݴٴ͢ΔtweetͷΈ • ߹ܭ71,315 tweetΛऩू 8 (Average)
ࢿՈͷߏ • ߹ܭ78,460ਓͷग़ࢿऀ ! • 4ճະຬͷࢿΛͨ͠ਓ51% • ؾ·͙Εࢿऀ “Occasional Investors”
• 32ճҎ্ͷࢿΛͨ͠ਓ11% • ৗ࿈ࢿऀ “Frequent Investors” ؾ·͙ΕࢿՈ ৗ࿈ࢿՈ 9
ϓϩδΣΫτͷΧςΰϦʔ͝ͱͷࢿऀͷ༁ • Music, DanceͳͲ୯ൃͷग़ࢿ͕ଟ͍ • Gamesৗ࿈ࢿՈ͕ଟ͍ˠେنͳήʔϜ։ൃͷืूͳͲ 10
ࢿՈͷڍಈʹؔ͢ΔԾઆ • ৗ࿈ࢿՈҎԼͷΑ͏ͳੑ࣭ͷϓϩδΣΫτʹࢿ͍͢͠ • ใ͕සൟʹΞοϓσʔτ͞ΕΔ • ϑΝϯμʔ͕ࢿՈͷ࣭ʹ͑Δ • ࢿͷใु͕ྑ͍ •
ߴ͍ࢿֹۚͷϓϩδΣΫτৗ࿈ࢿՈʹࢿ͞Ε͍͢ • ϩʔΧϧͳϓϩδΣΫτؾ·͙ΕࢿՈʹࢿ͞Ε͍͢ • ૣ͘ࢿΛूΊΔϓϩδΣΫτৗ࿈ࢿՈʹࢿ͞Ε͍͢ • ৗ࿈ࢿՈࣗͷڵຯ͋ΔϓϩδΣΫτʹࢿ͍͢͠ 11
ϓϩδΣΫτͰ͢Δಛྔ • ϓϩδΣΫτͷߋ৽ • ϑΝϯμʔͷίϝϯτ • ใुͷϨϕϧ • ΣϒαΠτͷ༗ແ •
ඪֹۚ ($) • ཧతͳڑͷΒ͖ͭ • ϓϩδΣΫτͷ 12
ͦΕͧΕͷಛྔ͝ͱͷࢿऀͷ༁ ϓϩδΣΫτͷߋ৽ ϑΝϯμʔͷίϝϯτ ใुͷϨϕϧ 13 ಛྔͷ͕૿͑Δ΄Ͳʹৗ࿈ࢿՈͷׂ߹͕૿Ճ͢Δ
ͦΕͧΕͷಛྔ͝ͱͷࢿऀͷ༁ (cont’d) ඪֹۚ ($) ཧతͳڑͷΒ͖ͭ ϓϩδΣΫτͷ 14 ඪֹ͕ۚ૿Ճ͢Δ΄Ͳʹ ؾ·͙ΕࢿՈͷׂ߹͕ݮগ͢Δ ؾ·͙ΕࢿՈͷࢿ
ʹӨڹ͞Εͳ͍
ԾઆɿࢿՈͷڵຯͱࢿઌͷؔ • LDA (Latent Dirichlet Allocation) ΛͬͨτϐοΫͷྨࣅ • ࢿͨ͠ϓϩδΣΫτͷ֓ཁͱࢿՈͷTweetͷ༰ (200
tweetsఔ) • (τϐοΫṖ) ! • ৗ࿈ࢿՈࣗͷڵຯ͋ΔෳͷτϐοΫʹࢿ͕ͪ͠ • ؾ·͙ΕࢿՈࣗͷڵຯͱؔͳ͍τϐοΫʹࢿ or ͻͱͭͷτ ϐοΫʹภͬͯࢿ͍ͯ͠Δ 15
͜͜·Ͱͷ·ͱΊ • ৗ࿈ࢿՈ (4ϲ݄ؒʹ32ճҎ্ࢿͨ͠Frequent Investors) • Α͘Ϛωʔδϝϯτ͞Εɼඪֹ͕ۚߴ͘ɼࣗͷڵຯʹ߹͏ϓϩδΣ Ϋτʹࢿ͢Δʹ͋Δ • ௨ৗͷࢿՈͷΑ͏ͳڍಈΛࣔ͢
• ؾ·͙ΕࢿՈ (4ϲ݄ؒʹ4ճະຬͷࢿΛͨ͠Occasional Investors) • ࠓճબΜͩಛྔʹ͋·ΓӨڹ͞ΕͣࢿΛߦ͏ • ࢿͱ͍͏ΑΓدͱ͍͏ײ͡ 16
ืूऀͷ༑ୡ͕ଟ͍΄Ͳ୯ൃͷग़ࢿׂ߹͕૿͑Δ • ϑΝϯμʔͷFacebookͷ༑ୡͷͱ ࢿՈͷ༁ͷؔ • ҼՌؔෆ໌ ! • Facebookͷ༑ୡͷ͕ଟ͍΄Ͳؾ·͙ ΕࢿՈΛूΊ͍͢
• Facebookͷ༑ୡͷ͕গͳ͍΄Ͳৗ࿈ ࢿՈΛूΊ͍͢…??? 17
ϑΝϯμʔͱग़ࢿऀͷϚονϯά Recommending Investors 18
ϓϩδΣΫτͱࢿՈͷϚονϯάํ๏ • Twitterʹ͍ΔજࡏతͳࢿՈʹରͯ͠ϓϩδΣΫτΛਪન͢Δ • KickstarterͷϢʔβ໊͔ΒTwitterͷΞΧϯτΛඥ͚ • 7,429ਓͷࢿՈ͕891ͷϓϩδΣΫτʹࢿͨ͠σʔλΛݩʹਪન • ࢿऀ͕ࢿ͢Δ= 1ɼࢿऀ͕ࢿ͠ͳ͍
= 0ͱͨ͠ೋྨ ! • Ϋϩʔϧͨ͠σʔλਖ਼ྫͷΈͳͷͰɼϥϯμϜͰෛྫΛࠞͥΔ • ਖ਼ྫෛྫͷׂ߹50-50 19
Ϩίϝϯσʔγϣϯͷख๏ͱධՁํ๏ • ੑೳධՁ͢Δख๏4ͭ : {LR, SVM-linear, SVM-poly, SVM-RBF} • ϩδεςΟοΫճؼʢLRʣ
• 3छྨͷΧʔωϧΛ༻͍ͨSVM ʢLinear, polynomial, RBFʣ ! • ධՁɿ5-fold cross validation • σʔληοτͷ80%Ͱֶशˠ20%ͰධՁ Λ5ճ܁Γฦͯ͠ධՁΛฏۉ 20
༻͢Δಛྔ • Static Feature: ϓϩδΣΫτൃ࣌ʹΘ͔Δಛྔ • ඪֹۚɾใुͷϨϕϧɾաڈʹࢿͨ͠ϓϩδΣΫτͷΧςΰϦɾ TwitterͷߘʹΑΔࢿՈͷڵຯ • Dynamic
Feature: ϓϩδΣΫτͷਐߦʹΑͬͯ໌ͯ͘͠Δಛྔ • ϓϩδΣΫτͷޭɾߋ৽ɾίϝϯτɾཧతͳڑͷΒ͖ͭ 21
ϨίϝϯσʔγϣϯͷධՁ • RBFΧʔωϧΛ༻͍ͨSVM • Static͚ͩͷಛྔ ɿ82% • Dynamic͚ͩͷಛྔɿ73% ! •
StaticͱDynamicΛ߹ΘͤΔͱACC 84% ! • ਖ਼ྫෛྫͷׂ߹͕50-50ɿϕʔεϥΠϯ50% ACC : Accuracy P : Precision R : Recall F1 : F-score AUC : ROCۂઢԼͷ໘ੵ 22
Ͳͷಛྔ͕ޮ͍͍ͯΔͷ͔ʁ • C : ίϝϯτ • R : ใुͷϨϕϧ •
S : ཧతͳڑͷΒ͖ͭ • G: • E : ΧςΰϦʔͷҰக • TS: ڵຯ͋ΔτϐοΫͱͷྨࣅ →EͱTS͕ਫ਼্ʹد༩͍ͯ͠Δ 23
จͷ·ͱΊ • ࢿՈʹΑͬͯKickstarterͷࢿελΠϧ͕ҧ͏ • ৗ࿈ࢿՈ৺తͳϓϩδΣΫτʹࢿ͢Δ • ؾ·͙ΕࢿՈدײ֮Ͱࢿ / ܳज़ʹؔ࿈ͨ͠ϓϩδΣΫτʹࢿ •
ࢿՈͱϓϩδΣΫτͷϚονϯάՄೳ • ࢿՈ͕ϓϩδΣΫτʹࢿΛ͢Δ͔Ͳ͏͔84%ͷਫ਼ͰਪଌՄೳ • ࢿՈͷڵຯ͋ΔΧςΰϦʔ༰͕Ϛονϯάʹڧ͘Өڹ͢Δ 24