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
Elasticsearchによる 全文検索の実装 in Rails
Search
FUJI Goro
March 24, 2018
Technology
6
9.5k
Elasticsearchによる 全文検索の実装 in Rails
Rails Developers Meetup 2018 (#railsdm) の発表資料です。
FUJI Goro
March 24, 2018
Tweet
Share
More Decks by FUJI Goro
See All by FUJI Goro
How to Boost Your Code with WebAssembly
gfx
2
2.9k
AssemblyScriptでライブラリコードの高速化をしてみる
gfx
5
3.2k
実践TypeScriptトークバトル
gfx
1
1.2k
歴史的経緯の説明 as code
gfx
7
2.8k
すばらしきGraphQLのSEKAIへようこそ
gfx
20
9.4k
マルチテナント・ウェブアプリケーションの実践
gfx
14
9.6k
How to choose the ORM on Android
gfx
1
4.3k
How Do We Get Along With Static Types
gfx
5
3.4k
"OSSにコントリビュート" なんてしてる場合じゃない!
gfx
21
15k
Other Decks in Technology
See All in Technology
BEYOND THE RAG🚀 ~とりあえずRAG?を超えていけ! 本当に使えるAIエージェント&生成AIプロダクトを目指して~ / BEYOND-THE-RAG-Toward Practical-GenerativeAI-Products-AOAI-DevDay-2025
jnymyk
4
520
Shadow DOM & Security - Exploring the boundary between light and shadow
masatokinugawa
0
690
ecspressoの設計思想に至る道 / sekkeinight2025
fujiwara3
12
1.8k
Amazon CloudWatchのメトリクスインターバルについて / Metrics interval matters
ymotongpoo
3
220
Semantic Machine Intelligence for Vision, Language, and Actions
keio_smilab
PRO
2
400
東京海上日動におけるセキュアな開発プロセスの取り組み
miyabit
0
160
みんな Kiro ってる?
r3_yamauchi
PRO
0
110
なぜAI時代に 「イベント」を中心に考えるのか? / Why focus on "events" in the age of AI?
ytake
2
690
「現場で活躍するAIエージェント」を実現するチームと開発プロセス
tkikuchi1002
6
1.1k
Kiro Hookを Terraformで検証
ao_inoue
0
110
Data Engineering Study#30 LT資料
tetsuroito
1
600
M365アカウント侵害時の初動対応
lhazy
7
4.9k
Featured
See All Featured
Building Better People: How to give real-time feedback that sticks.
wjessup
367
19k
The Invisible Side of Design
smashingmag
301
51k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
367
26k
Java REST API Framework Comparison - PWX 2021
mraible
31
8.7k
Thoughts on Productivity
jonyablonski
69
4.7k
Raft: Consensus for Rubyists
vanstee
140
7k
Faster Mobile Websites
deanohume
308
31k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
194
16k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
45
7.5k
Measuring & Analyzing Core Web Vitals
bluesmoon
7
530
GraphQLとの向き合い方2022年版
quramy
49
14k
How to Ace a Technical Interview
jacobian
278
23k
Transcript
ElasticsearchʹΑΔ શจݕࡧͷ࣮ in Rails #railsdm by FUJI Goro (gfx)
ࠓ͢͜ͱ • ΣϒαʔϏεͰશจݕࡧΛఏڙ͢Δͱ͖ʹ ͱ͖ʹߟ͑Δ͜ͱ • ElasticsearchʹΑΔશจݕࡧʹ͓͚Δ࣭ͷ ੍ޚʹ͍ͭͯ
ࠓ͞ͳ͍͜ͱ • Elasticsearch vs ଞͷશจݕࡧΤϯδϯ • ϚωʔδυαʔϏε vs ࣗલ •
ಛఆͷ elasticsearch client library ͷ • e.g. elasticsearch-rails, searchkick, elastomer-client
ൃදऀʹ͍ͭͯ • ໊લ: gfx / ౻ޗ • ใڞ༗αʔϏεKibelaΛ։ൃ͍ͯ͠Δ • Kibela
Ruby on RailsαʔϏε • લճͷrailsdmͰʮϚϧνςφϯτɾΣϒΞϓϦέʔ γϣϯͷ࣮ફʯͱ͍͏ൃදΛͨ͠ • KibelaͰ Amazon Elasticsearch Service Λར༻
ΣϒαʔϏεͰશจݕࡧΛ ఏڙ͢Δͱ͖ʹߟ͑Δ͜ͱ
શจݕࡧ in ΣϒαʔϏε • ࣭Λ্͛Δʹఔίετֻ͕͔Δ • ≒ίετΛֻ͚Δ΄Ͳʹ࣭Λ্͛ΒΕΔ • Ͳͷ͘Β͍ίετΛֻ͚ΒΕΔ͔ͰΔ͜ͱ͕ ͔ͳΓมΘΔ
• ίετΛֻ͚ΒΕͳ͍ͳΒͦͦશจݕࡧ ෆཁ͔͠Εͳ͍
ݕࡧͷ࣭ • ݕࡧͷ࣭Ұൠʹʮద߹ʯͱʮ࠶ݱʯ ͰݟΔ • ద߹ʮਖ਼֬ੑʯͱ͍͏ • ࠶ݱʮཏੑʯͱ͍͏
ద߹ vs ࠶ݱ • ద߹:ʮݕࡧ͞Εͨద߹จॻͷ ÷ ݕࡧ݁Ռͷจॻͷ ʯ • ద߹͕ߴ͍ͱʮݕࡧ݁Ռͷଟ͕͘ద߹ੑ͕ߴ͍ʯ
• ࠶ݱ:ʮݕࡧ͞Εͨద߹จॻͷ ÷ શରจॻதͷਖ਼ ղจॻͷʯ • ࠶ݱ͕ߴ͍ͱʮద߹ੑͱ͔͘औΓ͜΅͠ͳ͍ʯ
ద߹ͱ࠶ݱͷࢉग़ • ԿΛ࣋ͬͯʮద߹ʯͱ͢Δ͔Ϣʔβʔͷ಄ͷதʹ ͷΈ͋Δ • ≒ ಉ͡ݕࡧΩʔϫʔυͰϢʔβʔ͝ͱʹٻΊΔ ͷҟͳΔՄೳੑ͕͋Δ • ࣮ࡍͷαʔϏεͰଌఆ͢Δ͜ͱෆՄೳ
• ⇢ϢʔβʔͷߦಈΛੳͯ͠ධՁ͢Δඞཁ͕͋Δ
ద߹ͱ࠶ݱͰॏࢹ͢Δͷʁ • Ұൠʹద߹ͱ࠶ݱτϨʔυΦϑͷ͕ؔ͋ Δͱ͍ΘΕΔ • ͲͪΒ͕ॏཁ͔αʔϏεͷੑ࣭ʹΑΔ • KibelaͷΑ͏ͳاۀͷจॻݕࡧͰʮऔΓ͜΅ ͠ ≒
όάʯͱೝࣝ͞ΕΔ • ͭ·Γ࠶ݱʮ࠷ݶकΔ͖࣭ʯͰ͋Δ
ElasticsearchʹΑΔશจݕࡧ ʹ͓͚ΔݕࡧείΞͷߟ͑ํ
Elasticsearchͱ • શจݕࡧΤϯδϯ • SQLͷLIKEݕࡧͱͷҧ͍: • ܗଶૉղੳޙͷ୯ޠͰݕࡧͰ͖Δ • ਖ਼نԽ͞Εͨ୯ޠͰݕࡧͰ͖Δ •
ෳࡶͳείΞʹجͮ͘ιʔτ͕Ͱ͖Δ
Elasicsearch༻ޠ RDBMS Table Record Column Elasticsearch Index Document Field ※
Elasticsearch v5·Ͱ table = type ͱઆ໌͞Ε͖͕ͯͨ v6 Ͱtypeඇྭɺ v7 Ͱআ͞ΕΔͷͰҙ
ܗଶૉղੳ
ܗଶૉղੳͯ͠ݕࡧ • શจݕࡧͷߴԽͷͨΊʹసஔΠϯσΫεͳͲͷ ΞϧΰϦζϜΛ͍͍ͨ • సஔΠϯσΫεͷͨΊʹೖྗ͞ΕΔจষΛదͳ ୯ҐͰׂ͢Δඞཁ͕͋Δ • ຊޠͷΑ͏ʹ୯ޠͷ۠Γ͕͖ͬΓ͠ͳ͍ݴ ޠͷ߹N
Gram and/or ܗଶૉղੳͰׂ͢Δ
N Gram • N௨ৗ2͔3 • 2 gram ͷ߹ɺʮ͜ΜʹͪʯΛ • [“͜Μ”,
“Μʹ”, “ʹͪ”, “ͪ”] ʹ͚Δ
ܗଶૉղੳ • ʮ౦ژʹҾͬӽͨ͠ʯΛ • [“౦ژ”, “ʹ”, “Ҿͬӽ͠”, “ͨ”] ͳͲʹׂ ͢Δ
• ࢺʢ໊ࢺ, ܗ༰ࢺʣใ͕ར༻Ͱ͖ΔͷͰɺ ޙͷਖ਼نԽͰͦͷใΛར༻
LIKEݕࡧͱͷࠩ: ܗଶૉղੳฤ • “౦ژʹҾͬӽͨ͠” LIKE “%ژ%” • Ϛον͢Δ • [“౦ژ”,
“ʹ”, “Ҿͬӽ͠”, “ͨ”].include?(“ژ ”) • Ϛον͠ͳ͍
ܗଶૉղੳͱN Gramͷࠩ • ͨͱ͑ʮژʯͰʮ౦ژʯΛؚΉจॻΛ… • ܗଶૉղੳϕʔεͩͱݕࡧͰ͖ͳ͍ • ͭ·Γద߹ͷ໘Ͱ༏Ε͍ͯΔ • N
GramϕʔεͩͱݕࡧͰ͖Δ • ͭ·Γ࠶ݱͷ໘Ͱ༏Ε͍ͯΔ
ਖ਼نԽ
ͳͥਖ਼نԽ͕ඞཁ͔ • ܗଶૉղੳʴసஔΠϯσΫεʹΑΔݕࡧͷࣜ Խ: [“Ҿͬӽ͠”, “ͨ”].include?(“Ҿͬӽ͠”) • “Ҿӽ”, “Ҿӽ͠” ಉ༷ʹݕࡧ͍ͨ͠
⇢ ਖ਼نԽ
ܗଶૉղੳͱਖ਼نԽ • ܗଶૉղੳث͕ࣙॻͱࢺใΛͬͯਖ਼نԽ͢Δ • Elasticsearchͷ߹kuromoji plugin͕ຊޠͷܗଶૉղੳ ͱਖ਼نԽΛαϙʔτ • kuromoji_baseform: “ҿΈ”
⇢ “ҿΉ” • kuromoji_stemmer: “αʔόʔ” ⇢ “αʔό” • kuromoji_number: “ඦԁ” ⇢ “100ԁ”
είΞϦϯά
ݕࡧॱͷॏཁੑ • ͨͱ͑ɿʰژʱͰʰ౦ژʱͱ͍͏จॻ ώοτͯ͠΄͍͕͠ɺʰژʱͰώοτ͢ Δจॻ͕͕͋ΔͳΒͦͪΒΛ্Ґʹ͍࣋ͬͯ ͖͍ͨ • ͭ·Γߴ͍࠶ݱΛҡ࣋ͭͭ͠ݕࡧ݁Ռͷ্ ҐʮΑΓద߹͢ΔͱࢥΘΕΔʯจॻʹͨ͠ ͍
୯ޠׂ๏ʹΑΔॏΈ͚ • ܗଶૉղੳͰׂ͢Δͱద߹ߴ͘ͳΓ͕ ͕ͪͩ࠶ݱ͘ͳΓ͕ͪ • N GramͰׂ͢Δͱ࠶ݱߴ͘ͳΓ͕ͪͩ ͕ద߹͘ͳΓ͕ͪ • ⇢είΞϦϯάͷࡍʹॏΈ͚Λ͢ΕΑ͍
ElasticsearchͰͷ࣮ํ๏ • elasticsearch-model Ͱͷσʔλιʔεఆٛ UJUMFLVSPNPKJͰ ׂ͢Δ UJUMF@OHSBN/ (SBNͰׂ͢Δ
ϑΟʔϧυ͝ͱͷॏΈ͚ • ElasticsearchͰϑΟʔϧυ͝ͱʹॏΈ͚ ΛઃఆͰ͖Δ
ͦͷଞͷείΞϦϯάख๏ • Phrase Match / ϑϨʔζϚον • Field Value Factor
/ ϑΟʔϧυ • Decay / είΞͷݮਰ
Phrase Match • TwitterݕࡧGoogleݕࡧͰμϒϧΫΥʔτͰғΜͩͱ͖ʹ ʮϑϨʔζͦͷͷʯ͕ݕࡧରʹͳΔͷ͕Phrase Match • Kibelaͷ߹ɺϑϨʔζʹϚονͨ͠ΒείΞΛϒʔετ ͍ͯ͠Δ •
ྫ:ʮঁਆసੜʯͰݕࡧ͢ΔͱʮਅɾঁਆసੜIII NOCTURNEʯͷ΄͏͕ʮҟੈքʹঁਆͱͯ͠సੜ͠ ͨʯΑΓείΞ͕ߴ͘ͳΔ
Field Value Factor • ಛఆͷϑΟʔϧυͷͰείΞΛϒʔετͤ͞ Δػೳ • Kibelaͷ߹ɺϥΠΫͱτϥοΫόοΫΛ Field Value
Factorͱͯ͠ө͍ͤͯ͞Δ • ͭ·ΓʮΑ͘ϥΠΫ͞ΕΔจॻʯʮΑ͘Ϧ ϯΫ͞ΕΔจॻʯͷείΞ͕ߴ͍
Decay • Field Value FactorͷٯͰɺಛఆϑΟʔϧυ ͰείΞΛݮਰͤ͞Δػೳ • KibelaͰɺจॻͷλΠϜελϯϓΛར༻͠ ͕ͯ࣌ؒܦͭ΄ͲʹείΞ͕ݮਰ͢ΔΑ͏ʹ ͍ͯ͠Δ
DecayͷΠϝʔδ • origin = now / timestampաڈͷҰͱ͢Δ
·ͱΊ • શจݕࡧͷ࣭ʮద߹ʯʮ࠶ݱʯʮε ίΞϦϯάʯͰߟ͑Δ • ʮద߹ʯͱʮ࠶ݱʯ୯ޠׂ๏ʹΑͬ ͯվળͰ͖Δ • ʮείΞϦϯάʯ༷ʑͳཁૉ͕͋ΔͷͰશ จݕࡧΤϯδϯͷυΩϡϝϯτΛख़ಡ͢Δ