Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
Temporal Knowledge Graphで作る! 時間変化するナレッジを扱うAI Ag...
Search
Sponsored
·
Ship Features Fearlessly
Turn features on and off without deploys. Used by thousands of Ruby developers.
→
po3rin
October 30, 2025
Programming
2.5k
7
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Temporal Knowledge Graphで作る! 時間変化するナレッジを扱うAI Agentの世界
po3rin
October 30, 2025
More Decks by po3rin
See All by po3rin
Go × SIMDで高速化するベクトル検索 ~ルーフラインモデルでSIMDが効く境界を探れ! ~
po3rin
1
6.4k
Driving AI Adoption Using In-House GPUs to Serve Qwen
po3rin
2
600
検索設計から 推論設計への重心移動と Recall-First Retrieval
po3rin
5
2.4k
LLMで複雑な検索条件アセットから脱却する!! 生成的検索インタフェースの設計論
po3rin
4
1.6k
C-Shared Buildで突破するAI Agent バックテストの壁
po3rin
0
560
嗚呼、当時の本番環境の状態で AI Agentを再評価したいなぁ...
po3rin
0
640
自作LLM Native GORM Pluginで実現する AI Agentバックテスト基盤構築
po3rin
2
1.3k
麻雀点数計算問題生成タスクから学ぶ Single Agentの限界と Agentic Workflowの底力
po3rin
5
3.4k
LLMOpsのパフォーマンスを支える技術と現場で実践した改善
po3rin
8
1.3k
Other Decks in Programming
See All in Programming
現場と育てるRailsモデリング 〜Spreadsheetから始めたモデル設計〜
siroemk
2
1.2k
ChromeでローカルLLM
ryu1013
0
190
No More NoSuchMethodError
exoego
1
120
[DroidKnights 2026] Workflow AX: From Auto-Documented Policies to E2E Testing
raeminkang
0
310
そのデザイントークン、Flutterでも同じ見た目になりますか?
nakamizu
0
240
作れる、でも消せるか?──安全に「枯らして消す」機能削除術 / Drain, Then Delete
chobishiba
3
1.3k
FactoryBotをやめればRSpecは速くなるのか?
yuta_onishi_97
0
1.2k
React Native에 현대적인 빌드 도구 더하기: Rollipop 구현기
leegeunhyeok
0
180
S3 TablesのIcebergデータに、AIエージェントからアクセスしてみた / 20261016developersio2026joetsu-accessing-iceberg-in-s3tables-via-agent
kasacchiful
PRO
0
260
Inertia Rails: The One-Person Framework with React & Friends
skryukov
0
210
sbt 2.0: takeoff
eed3si9n
0
190
선언형 햅틱 라이브러리 진동을 소개합니다
l2hyunwoo
1
250
Featured
See All Featured
GraphQLの誤解/rethinking-graphql
sonatard
75
12k
How to Build an AI Search Optimization Roadmap - Criteria and Steps to Take #SEOIRL
aleyda
1
2.2k
Responsive Adventures: Dirty Tricks From The Dark Corners of Front-End
smashingmag
254
23k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
47
8.4k
Darren the Foodie - Storyboard
khoart
PRO
4
4.1k
Rails Girls Zürich Keynote
gr2m
96
14k
Introduction to Domain-Driven Design and Collaborative software design
baasie
1
1.1k
Prompt Engineering for Job Search
mfonobong
0
570
Dominate Local Search Results - an insider guide to GBP, reviews, and Local SEO
greggifford
PRO
0
390
The Curse of the Amulet
leimatthew05
3
16k
How to Grow Your eCommerce with AI & Automation
katarinadahlin
PRO
2
340
Building Better People: How to give real-time feedback that sticks.
wjessup
371
20k
Transcript
5FNQPSBM,OPXMFEHF(SBQIͰ࡞Δʂ ࣌ؒมԽ͢ΔφϨοδΛѻ͏"*"HFOUͷੈք )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
)JSPNV/BLBNVSB )JSPNV /BLBNVSB QPSJO .-0QT&OHJOFFS*OUFSFTUJOH JO*OGPSNBUJPO3FUSJFWBM
-BZFS9ݱࡏ"*"HFOUΛ ΰϦΰϦਐΊ͍ͯΔ https://layerx.co.jp/news/20250401/ 1
"*"HFOU༻ͷϝϞϦφϨοδΛ࣋ͭඞཁੑ 2 )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO ϝϞϦΛόοΫΤϯυͷͰͳ͘ɺͱͯ͠ ѻ͏͖Ͱ͢ɻࠓɺϝϞϦೝࣝΛج൫ʹߏங͢Δ ελʔτΞοϓ͕ͦ͜ɺະདྷͷ࠷ΠϯςϦδΣϯ τͰύʔιφϥΠζ͞Εɺ͔ͭڧྗͳAIγεςϜΛ ߏங͢ΔͰ͠ΐ͏ɻ(The State of
AI 2025ΑΓ) https://www.bvp.com/atlas/the-state-of-ai-2025
"*ਃϨϏϡʔ 3 )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO * ਃϧʔϧձࣾʹΑͬͯશ͘ҧ͏ͷͰ͋Γɺঘ ͭͦͷϧʔϧ࣌ؒܦաͰมԽ͠·͢ɻ * ϧʔϧΛφϨοδͱͯ͠දݱ͢ΔͨΊʹ࣌ؒܦ աʹΑΔߋ৽ΛදݱͰ͖ͳ͍͚ͯ͘·ͤΜɻ *
·ͨɺจষԽ͞Ε͍ͯͳ͍ϧʔϧͷ߹ɺ҉తʹ ӡ༻͞Ε͍ͯΔϧʔϧΛφϨοδͱͯ͠ಈతʹΈ ࠐΜͰ͍͘ඞཁ͕͋Γ·͢ɻ * ͜ΕΒͷ՝ΛࠓճTemporal Knowledge GraphͰ Ͳ͏ಥഁͰ͖Δͷ͔Λ͠·͢ɻ
ͳͥ,OPXMFEHF(SBQI͕͞Ε͖͔ͯͨ 4 * ίϯςΩετ/ϝϞϦͱͯ͠ɺෳࡶͳࣝମܥ ΛදݱͰ͖Δʂʂ(ෳͷ֓೦ؒͷෳࡶͳؔ ੑਂ͍จ຺ཧղΛඞཁͱ͢ΔλεΫ) * LLMͷొͰKnowledge Graph͕ߏங͘͢͠ ͳͬͨʂʂ
* ྫ͑ɺGraphRAGͷจͰɺLLMϕʔε ͷΤϯςΟςΟɺϦϨʔγϣϯͷநग़Λߦ ͍ͬͯ·͢ɻ https://arxiv.org/abs/2404.16130 )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
(SBQI3"(ͷऑ 5 GraphRAGࣄ࣮͕࣌ؒͱͱʹมԽ͠ͳ͍੩తͳKnowledge GraphʹযΛ͓ͯͯΓɺ࣌ؒมԽʹ͏ಈతͳਐԽΛແࢹ ͍ͯ͠Δɻɻɻ )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
5FNQPSBM,OPXMFEHF(SBQIʂʂ 6 Temporal Knowledge Graph(TKG)࣌ؒೝࣝܕͰಈతͳมߋ ʹڧ͍Knowledge GraphͰ͢ɻ ձσʔλͷதͰग़ݱ͢ΔϢʔβʔͷ৽͍͠ᅂͪΖΜɺ ࣌ؒܦաͰՃ͞ΕΔܾ࣍ࢉॻࣾϧʔϧͳͲɺ࣌ؒܦա ͰมԽ͢ΔφϨοδΛѻ͏͜ͱ͕Ͱ͖·͢ɻ
ϢʔβʔͱͷΓऔΓͰAI Agentͷߦಈ͕ஞ࣍తʹվળ͞Εͯ ͍͘αΠΫϧΛճ͢͜ͱ͕Ͱ͖·͢ɻ )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
Graphiti(1) * GraphitiTKGͷҰछͰ͋ΓɺͦΕΛϝϞϦϓ ϥοτϑΥʔϜͱͯ͠ѻ͏Zep͕͋Γ·͢ * GraphitiOSSɺZepZep Cloudͱͯ͠༗ঈར ༻ * Zep(Graphiti)Preston
RasmussenΒͷจͰ հ͞Ε͍ͯ·͢ɻ )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO https://arxiv.org/abs/2501.13956 7
(SBQIJUJ 8 * LLMGraphiti3ஈͷαϒάϥϑͰߏ͞Ε·͢ɻ * Τϐιʔυαϒάϥϑ: ੜͷೖྗσʔλ(ϝοηʔδͳͲ)Λଛࣦͳ͠Ͱอଘ͠ɺ࣌ؒతͳॱং Λҡ࣋͢ΔɻηϚϯςΟοΫΤϯςΟςΟؔੑΛநग़͢Δࡍͷج൫ͱͳΔɻ *
ηϚϯςΟοΫΤϯςΟςΟαϒάϥϑ: Τϐιʔυαϒάϥϑ͔Βநग़͞ΕͨΤϯςΟς Οʢਓɺɺॴɺ֓೦ͳͲʣͱɺͦΕΒͷؒͷؔੑΛߏԽ͞ΕͨܗͰදݱ͢Δɻ * ίϛϡχςΟαϒάϥϑ: ηϚϯςΟοΫΤϯςΟςΟͷΫϥελʔʢίϛϡχςΟʣΛܗ ͠ɺ͜ΕΒͷΫϥελʔͷߴϨϕϧͳཁΛఏڙ͢Δɻ )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
(SBQIJUJ 9 * ݕࡧ࣍ͷࡾͭΛ࣮ߦͯ͠ϦϥϯΫ * 1. ϕΫτϧݕࡧ(Cosine Similarity) *
ΫΤϦςΩετ͕ຒΊࠐΈϕΫτϧʹม͞ΕɺίαΠϯྨࣅͰݕࡧ͞Ε·͢ɻ * 2. จࣈྻݕࡧ(BM25/Fulltext) * ಉ͡ΫΤϦςΩετ͕BM25ΞϧΰϦζϜΛͬͨϑϧςΩετݕࡧʹΘΕ·͢ * 3. άϥϑݕࡧ(BFS) * BFS(෯༏ઌ୳ࡧ)ͰάϥϑΛτϥόʔαϧ͠·͢ɻॳճͷϕΫτϧݕࡧɾจࣈྻݕࡧͷ݁Ռ͔Βى ϊʔυΛऔಘ͠ɺ͔ͦ͜Βؔ࿈͢ΔΤοδϊʔυΛ୳ࡧ͠·͢ɻ * ݕࡧ݁Ռʹରͯ͠Reciprocal Rank Fusion (RRF) Maximal Marginal Relevance (MMR) ͳͲͷطଘͷϦ ϥϯΩϯάΞϓϩʔνΛαϙʔτ͍ͯ͠·͢ɻ )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
10
12 5,(Λ͏1P$σϞ ਓؒʹΑΔࠩ͠͠ίϝϯτ͔Β ֶश͠ਃΛϨϏϡʔ͢ΔAI Agentػೳ
࣮ 13
͔ͬͨ͠ͱ͜Ζɺ·͍ͩ͠ͱ͜Ζ 14 * ΓKnowledge Graphߏங෦ɻ * νϟϯΫԽɻ͋·Γʹ͍จষΛLLMʹ༩͑Δͱநग़͕͏·͍͖͘·ͤΜɻͦ͜ͰจॻΛҰ ߦ୯ҐͷΤϐιʔυܗࣜʹมͯ͢͠ * ຊޠͰ՝ͱͳΔͷʮओޠͷলུʯͰ͢ɻ͜͜ϓϩϯϓτͰ໌֬ʹରԠ͢Δඞཁ͕͋
Δɻ * Temporal Knowledge GraphͰ࣌ؒΛ༩্ͨ͠ͰదʹφϨοδΛ্ॻ͖͍ͨͨ͠Ίɺ ಉ͡ΤϯςΟςΟΛผͷΤϯςΟςΟͱͯ͠ѻͬͯ͠·͏ͷக໋తɻ * LLMʹΑΔΤϯςΟςΟղܾ͕༳ΕΔ߹ɺϊʔυॏෳഉআɺΤοδॏෳഉআϩδοΫΛ ผͰߟ͑Δඞཁ͕͋Γ·͢ɻ͜ͷ՝ͷରॲݱࡏ୳ٻதɻ )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
·ͱΊ 15 * Temporal Knowledge Graph࣌ؒతͳφϨοδͷมԽΛଊ͑Δ͜ͱ͕Ͱ͖Δɻ * ࣌ؒೝࣝܕφϨοδݕࡧ҉తʹมԽ͍ͯ͘͠φϨοδΛѻ͏υϝΠϯͰ༗ޮ * GraphitiͰϙϯͰ·ͩ͏·͘ಈ͔ͳ͍ɻগ͠ίπ͕͍Δɻ
)JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO
ϒϩάʹৄࡉΛެ։ͯ͠·͢ 16 )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO https://tech.layerx.co.jp/entry/tkg-agent ࠓͷҎ֎ʹ… * Knowledge Graph͕༗ޮͳλεΫͱʁ * GraphitiҎ֎ͷTKGจͷհ
* φϨοδάϥϑͷධՁ
5FNQPSBM,OPXMFEHF(SBQIͰ࡞Δʂ ࣌ؒมԽ͢ΔφϨοδΛѻ͏"*"HFOUͷੈք )JSPNV/BLBNVSB4FBSDI&OHJOFFSJOH5FDI5BML"VUVNO