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
Contextual Bandit による表示コンテンツの最適化
Search
j.harashima
October 16, 2014
3
3.2k
Contextual Bandit による表示コンテンツの最適化
j.harashima
October 16, 2014
Tweet
Share
More Decks by j.harashima
See All by j.harashima
日本語レシピデータセットの継続的な構築と複合的な利用/JED2022
junharashima
0
12k
自然言語処理の基礎と応用 〜 料理と医療を題材として 〜 /JADI2021
junharashima
0
4.7k
企業での研究開発の楽しさと苦労/WAP-Tech-Talk
junharashima
1
11k
クックパッドにおける研究開発/HCG2020
junharashima
0
4k
Calorie Estimation in a Real-World Recipe Service/iaai-20
junharashima
0
14k
クックパッドと機械学習(短縮版)/MLTBP
junharashima
0
9k
クックパッドと NLP/CV/nlpaper-challenge
junharashima
0
110
AI 部門の実戦投入/repro-tech-meetup
junharashima
0
4k
クックパッドと機械学習/uec-aix-seminar
junharashima
0
11k
Featured
See All Featured
Raft: Consensus for Rubyists
vanstee
140
7.1k
Statistics for Hackers
jakevdp
799
220k
What's in a price? How to price your products and services
michaelherold
246
12k
GraphQLとの向き合い方2022年版
quramy
49
14k
Producing Creativity
orderedlist
PRO
347
40k
Testing 201, or: Great Expectations
jmmastey
45
7.6k
Speed Design
sergeychernyshev
32
1.1k
How to Ace a Technical Interview
jacobian
279
23k
Site-Speed That Sticks
csswizardry
10
790
How to Think Like a Performance Engineer
csswizardry
25
1.8k
Designing Dashboards & Data Visualisations in Web Apps
destraynor
231
53k
A designer walks into a library…
pauljervisheath
207
24k
Transcript
$POUFYUVBM#BOEJUʹΑΔ දࣔίϯςϯπͷ࠷దԽ ΫοΫύουגࣜձࣾ ٕज़෦ݕࡧάϧʔϓ ݪౡ७
ݪౡ७ wΤϯδχΞ!ΫοΫύου wത࢜ʢใֶʣ wࣗવݴޠॲཧˍใݕࡧ
ͷ݄ؒར༻ऀສਓ Ϩγϐສ ʢ݄ʣ
ϞόΠϧར༻ऀͷ૿Ճ
1$ εϚϗʢϒϥβʣ εϚϗʢΞϓϦʣ Ψϥέʔ
ը໘αΠζ ʙΠϯν ʙΠϯν ʙΠϯν
6*͕ෳࡶԽ ݙཱΛग़ͨ͠Βศརʁ Ϣʔβ͞ΜʹΑΔ ͓͢͢ΊϨγϐͱ͔ ྉཧͷίπͳΜͯͲ͏ʁ ৽͍͠ίϯςϯπΛग़ͦ͏ʂ χϡʔεɺಈըɺ JOݕࡧ݁Ռϖʔδ Ϩγϐ͚ͩͰϢʔβͷཁʹԠ͑ΒΕ ͳ͍͕ɺίϯςϯπΛେྔʹग़͢ͷ
౷߹ݕࡧ
ݕࡧ݁Ռϖʔδ ΫΤϦ͝ͱʹ ࠷దͳίϯςϯπ͚ͩΛදࣔ ʢ݄ʙɺࠓϒϥβ൛ͷΈରԠʣ Ͳ͏ͬͯʁ
.VMUJ"SNFE#BOEJU w ୳ࡧͱ׆༻Λ͍͚ͯใुΛ࠷େԽ w ୳ࡧɿใΛऩू͢ΔͨΊʹΛબ w ׆༻ɿऩूͨ͠ใʹج͍ͮͯΛબ w ୳ࡧͱ׆༻τϨʔυΦϑ
$POUFYUVBM#BOEJU w ͷใु͕ίϯςΩετʹΑͬͯҧ͏ w FHࠂͷ$53ݟΔਓʹΑͬͯҧ͏ w ίϯςΩετΛૉੑͰදݱ w ใु͕࠷େͱͳΔΑ͏ʹॏΈΛֶश w
ίϯςΩετ͝ͱʹ࠷దͳΛબ
#BOEJUJODPPLQBE w ɿίϯςϯπʢFHݙཱɺχϡʔεʣ w ใुɿ$53 w ૉੑɿΫΤϦͳͲ w ֶशɿ7PXQBM8BCCJUΛར༻ʢരʂʣ w
୳ࡧɿҰ෦17ͰϥϯμϜʹίϯςϯπΛදࣔ w ׆༻ɿ$53࠷େͱظ͞ΕΔίϯςϯπΛදࣔ ʢ,%%ͷ:BIPP-BCTͷจΛΞϨϯδʣ
શମਤ 0⒐JOF 0OMJOF explore exploit log*show/click*events Treasure*Data
FH݄ʹࡾͷσʔλͰֶशͯ͠ɺҰͷσʔλʹରͯ͠ਪఆ͢Δ߹ 0⒐JOF 0OMJOF 10/9 10/10% 10/11% 10/9 10/10% 10/11% m/d
m/d md% % md% % construct%training% and%test%data m/d% % %m’/d’% mdm’d’ % % Amazon%S3 10/9% % 10/11% … … predict%ac:ons learn%model% % % Treasure%Data
ิɿ୳ࡧͷׂ߹ w ݱࡏ w ҰఆʢҰ෦17ʣ w ࠓޙ w ঃʑʹσʔλ͕ੵʢJF৴པ্͕ঢʣ w
୳ࡧͷׂ߹ݮΒ͍͚ͯ͠Δ
ิɿਪఆͷํ๏ w ϦΫΤετͷʹਪఆ w ϨεϙϯελΠϜ͕ѱԽͦ͠͏ w ϦΫΤετͷલʹਪఆ w ߴසͷίϯςΩετʹରͯ͠༧ΊΛ ਪఆɺ݁ՌΛςʔϒϧʹอ͍࣋ͯ͠Δ
݁Ռ w 6*Λγϯϓϧʹอͯͨ w $53͕ߴ͍ίϯςϯπΛ දࣔͰ͖ΔΑ͏ʹͳͬͨ ʢ୳ࡧ݁Ռͱൺͯʣ ʮΦϜϨπʯ ݙཱ ˞݄࣌Ͱͷ݁Ռ
ʮϋϯόʔάʯ ʮ͓ͰΜʯ ňೕ৯ʼn χϡʔε ΧϑΣʢܝࣔ൘ʣ ͓͢͢ΊϨγϐ
ৄࡉ ଓ͖8FCͰʂ ΫοΫύου։ൃऀϒϩά ݕࡧ ʢ݄ࠒɺଟʣ
·ͱΊ w ϞόΠϧར༻ऀͷ૿Ճ w ϞόΠϧը໘͕খ͍͞ w ίϯςϯπΛେྔʹදࣔˠ6*͕ෳࡶԽ w ΫΤϦ͝ͱʹ࠷దͳίϯςϯπ͚ͩΛදࣔ w
$POUFYUVBM#BOEJUͷར༻
ࠓޙͷల w ୳ࡧׂ߹ͷௐ w ৴པͷಋೖʢͬͱڧԽֶशͬΆ͘ʂʣ w ίϯςϯπͷՃ w ྉཧͷಈըɺྉཧͷίπɺ w
ΞϓϦͷల։