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
Go as an aggregator in recommendation systems
Search
Agata Naomichi
July 26, 2018
Programming
1.5k
2
Share
Go as an aggregator in recommendation systems
Agata Naomichi
July 26, 2018
More Decks by Agata Naomichi
See All by Agata Naomichi
Why Kotlin? 電子カルテを Kotlin で開発する理由 / Why Kotlin? at Henry
agatan
2
9.3k
全員アーキテクトで挑む、 巨大で高密度なドメインの紐解き方
agatan
8
23k
チームで開発し事業を加速するための"良い"設計の考え方 @ サポーターズCoLab 2025-07-08
agatan
2
690
医療系スタートアップが経験した 認知負荷問題の症状分析と処方箋 チーム分割による認知負荷の軽減 / Cognitive Load Busters
agatan
2
620
専門性の高い領域をいかに開発し、 テストするか / How to test and develop complicated systems with Domain Experts!
agatan
3
900
Henry のサーバーサイドアーキテクチャ 狙いと課題 2022.08.25 / Server-Side Architecture at Henry, Inc.
agatan
3
6k
The Web Conference 2020 - Participation Report
agatan
1
750
○○2vec 再考
agatan
1
4.7k
Improving "People You May Know" on Directed Social Graph
agatan
4
2.8k
Other Decks in Programming
See All in Programming
運用エージェントは "作る" から "育てる" へ - 記憶と自己進化の3層設計パターン / self-evolving-agents-three-layer-agent-design
gawa
11
2.6k
RTSPクライアントを自作してみた話
simotin13
0
150
Old Dog, New Tricks: The Java 25 Reinvention - JNation
bazlur_rahman
0
120
SPMマルチモジュールで テストカバレッジを取得する技法
yosshi4486
0
120
AIエージェントの隔離技術の徹底比較
kawayu
0
430
ReactとSvelteのその先、Ripple-TS / Beyond React and Svelte: Ripple-TS
ssssota
3
830
TypeSpec で繋ぐ複数プロダクトの型安全
maroon8021
1
240
Why Laravel apps break—Mastering the fundamentals to keep them maintainable
kentaroutakeda
1
200
Cloudflare で始める Data Platform
ta93abe
0
300
Talking to terminals (and how they talk back) (KotlinConf 2026)
jakewharton
PRO
1
120
次世代リンターで探る、tsgo 時代における型認識カスタムルールの現実解
ytakahashii
1
990
不変条件と整合性境界—ビジネスが決める設計判断と実現パターン / Invariants and Consistency Boundaries
nrslib
8
1.7k
Featured
See All Featured
A designer walks into a library…
pauljervisheath
211
24k
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
kimpetersen
PRO
0
340
Data-driven link building: lessons from a $708K investment (BrightonSEO talk)
szymonslowik
1
1.1k
The SEO identity crisis: Don't let AI make you average
varn
0
470
The AI Revolution Will Not Be Monopolized: How open-source beats economies of scale, even for LLMs
inesmontani
PRO
3
3.5k
GraphQLの誤解/rethinking-graphql
sonatard
75
12k
Money Talks: Using Revenue to Get Sh*t Done
nikkihalliwell
0
230
Groundhog Day: Seeking Process in Gaming for Health
codingconduct
0
190
How to build a perfect <img>
jonoalderson
1
5.5k
GitHub's CSS Performance
jonrohan
1033
470k
The Curious Case for Waylosing
cassininazir
1
360
Imperfection Machines: The Place of Print at Facebook
scottboms
270
14k
Transcript
©2018 Wantedly, Inc. Go as an aggregator In Recommendation Systems
26.Jul.2018 - Naomichi Agata
©2018 Wantedly, Inc. agatan Software engineer at Wantedly, inc. Server
side + Machine learning Github Twitter @agatan @agatan_
©2018 Wantedly, Inc. Everything is a Recommendation https://medium.com/netflix-techblog/netflix-recommendations-beyond-the-5-stars-part-1-55838468f429
©2018 Wantedly, Inc. Recommendations ΄ͱΜͲͷαʔϏεͰʮԿ͔Λਪન͢Δʯͱ͍͏ػೳ͋Δ
©2018 Wantedly, Inc. Impact of Recommendations ⾣Linkedinͷͭͳ͕Γͷ50%Ҏ্ʮΓ߹͍Ͱ͔͢ʁʯܦ༝ ⾣ https://engineering.linkedin.com/teams/data/projects/pymk ⾣NetflixਪનγεςϜͷcompetition
Λ։࠵ۚ͠$1 Million Λग़͍ͯ͠Δ ⾣ https://medium.com/netflix-techblog/netflix-recommendations-beyond-the-5-stars-part-1-55838468f429
©2018 Wantedly, Inc. Components of Recommendations ਪનͷࠜڌ͍ΖΜͳॴʹ͋Δ ⾣ʮ˓˓͞Μ͕-JLF͠·ͨ͠ʯ ⾣ʮ͜ͷΛߪೖͨ͠ਓ͜ͷങ͍ͬͯ·͢ʯ ⾣ʮڞ௨ͷͭͳ͕Γ͕ਓ͍·͢ʯ
⾣ʮͷχϡʔεʯ ⾣ʮ˓˓Λݕࡧͨ͠ํʯ ⾣ʮಉ͡ձࣾͰಇ͍͍ͯΔϢʔβʯ
©2018 Wantedly, Inc. Order of Recommendations ΑΓྑ͍ΞΠςϜΛɺΑΓྑ͍ॱংͰఏࣔ͢Δ͜ͱ͕ٻΊΒΕΔ ⾣Hot Topics ͳΔ͘͘ఏ͍ࣔͨ͠
⾣֬ݻͨΔࣗ৴ͷ͋ΔਪનΛ༏ઌͯ͠ݟ͍ͤͨ ⾣શମͰͷਓؾॱΑΓύʔιφϥΠζͨ͠ॱংΛఏڙ͍ͨ͠ ⾣αʔϏεݻ༗ͷׂΓࠐΈ͋Δ͔͠Εͳ͍ by Google
©2018 Wantedly, Inc. Recommendations with strategies
©2018 Wantedly, Inc. aggregator Strategy Strategy Strategy Strategy ༑ୡ͕-JLFͨ͠ΞΠςϜΛఏࣔ ߪೖཤྺ͔Βͷ͓͢͢Ί
ͷΞΠςϜ ϓϩϑΟʔϧ͔Βͷ͓͢͢Ί Recommendations with strategies Strategy Λ࣮ߦ ݁ՌΛू re-ordering ฒߦॲཧ
©2018 Wantedly, Inc. Recommendations with strategies ֤strategy͕ਪનΞΠςϜΛఏࣔ UZQF4USBUFHZJOUFSGBDF\ /BNF 4USBUFHZ/BNF
4VHHFTU DUYDPOUFYU$POUFYU VTFS*%JOU TJ[FJOU <> 4VHHFTU6TFS FSSPS ^ UZQF4VHHFTU6TFSTUSVDU\ 6TFS*%JOU 4DPSFqPBU 4USBUFHZ/BNF4USBUFHZ/BNF 3FBTPOJOUFSGBDF\^GPSMPHHJOH ^
©2018 Wantedly, Inc. Recommendations with strategies ͦΕΒΛฒߦʹ࣮ߦ͠ू ࠷ऴతͳॱংΛܾఆ͢Δ GPS@ TSBOHFTUSBUFHJFT\
XH"EE HPGVOD T4USBUFHZ \ EFGFSXH%POF TT FSST4VHHFTU DUY VTFS*% TJ[F JGFSSOJM\ SFQPSUFSSPS SFUVSO ^ NV-PDL TVHHFTUJPOTBQQFOE TVHHFTUJPOT TT NV6OMPDL ^ T ^
©2018 Wantedly, Inc. Why Go? Machine Learning ͱ Microservices ͳߏ
ˠ֤Strategy API CallΛؚΈ͏Δ ˠฒߦॲཧʹڧ͍͜ͱ͕׆͖Δ aggregator microservices
©2018 Wantedly, Inc. Responsibility of Aggregator ⾣Logging ⾣ ͲͷStrategy ͕Ͳͷ͘Β͍ՌΛ͍͋͛ͯΔ͔
⾣֤Strategy ͷείΞͷॏΈ͚ʹΑΔϥϯΩϯά ⾣e.g. ͢Ͱʹఏࣔͨ͜͠ͱͷ͋ΔΞΠςϜͷείΞΛݮਰͤ͞Δ ⾣A/B Testing
©2018 Wantedly, Inc. Problems… ⾣Frror reporting ⾣ ͋Δstrategy ͕ࢮΜͰ͍ͯશମࢭ·Βͳ͍Ͱ΄͍͠ ⾣ࢮΜͩ͜ͱʹؾ͖͍ͨ
⾣Frror reporting service Λ׆༻ ⾣ෳͷStrategy Ͱಉ͡API CallΛͨ͘͠ͳΔ ⾣e.g. Profile Service ʹॴଐΛ͍߹ΘͤΔ࠷ۙങͬͨͷΧςΰϦ͕Γ͍ͨ ⾣HPMBOHPSHYTZODTJOHMFqJHIU ΠϯϝϞϦΩϟογϡͰແཧཧଋͶΔ ⾣Ͳ͜·Ͱaggregator ͕ܭࢉ͢Δ͖͔
©2018 Wantedly, Inc. Conclusion ⾣ਪનγεςϜ͍ΖΜͳཁૉͷΈ߹Θͤ ⾣Microservices / Machine Learning ⾣A/B
Test ͕ॏཁ ⾣࣮ࡍʹԿΛݟ͔ͤͨɺԿ͕Action ʹͭͳ͕͔ͬͨLogging ͍ͨ͠ ⾣લஈʹGo Λ͓͘ͱศར ⾣Concurrent ʹ͍ΖΜͳStrategy Λ࣮ߦ͢Δͷ͕؆୯