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
論文紹介: Neural Relation Extraction for Knowledge ...
Search
Koji Matsuda
September 27, 2019
Research
0
410
論文紹介: Neural Relation Extraction for Knowledge Base Enrichment (ACL2019)
Koji Matsuda
September 27, 2019
Tweet
Share
More Decks by Koji Matsuda
See All by Koji Matsuda
Editing Factual Knowledge in Language Models (EMNLP 2021)
conditional
0
340
SHINRA2020-JP リーダーボードのご案内
conditional
0
160
AI王 〜クイズAI日本一決定戦〜
conditional
0
11k
Training Classifiers with Natural Language Explanations
conditional
0
300
Other Decks in Research
See All in Research
ナレッジプロデューサーとしてのミドルマネージャー支援 - MIMIGURI「知識創造室」の事例の考察 -
chiemitaki
0
220
研究を支える拡張性の高い ワークフローツールの提案 / Proposal of highly expandable workflow tools to support research
linyows
0
300
第79回 産総研人工知能セミナー 発表資料
agiats
3
200
情報処理学会関西支部2024年度定期講演会「自然言語処理と大規模言語モデルの基礎」
ksudoh
10
2.5k
地理空間情報と自然言語処理:「地球の歩き方旅行記データセット」の高付加価値化を通じて
hiroki13
1
190
o1 pro mode の調査レポート
smorce
0
110
Zipf 白色化:タイプとトークンの区別がもたらす良質な埋め込み空間と損失関数
eumesy
PRO
8
1.3k
知識強化言語モデルLUKE @ LUKEミートアップ
ikuyamada
0
200
LLM 시대의 Compliance: Safety & Security
huffon
0
590
Poster: Feasibility of Runtime-Neutral Wasm Instrumentation for Edge-Cloud Workload Handover
chikuwait
0
340
Data-centric AI勉強会 「ロボットにおけるData-centric AI」
haraduka
0
440
Geospecific View Generation - Geometry-Context Aware High-resolution Ground View Inference from Satellite Views
satai
2
240
Featured
See All Featured
Code Reviewing Like a Champion
maltzj
521
39k
Fashionably flexible responsive web design (full day workshop)
malarkey
406
66k
The Power of CSS Pseudo Elements
geoffreycrofte
75
5.5k
Being A Developer After 40
akosma
89
590k
Rebuilding a faster, lazier Slack
samanthasiow
80
8.8k
Agile that works and the tools we love
rasmusluckow
328
21k
Scaling GitHub
holman
459
140k
Practical Orchestrator
shlominoach
186
10k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
193
16k
How To Stay Up To Date on Web Technology
chriscoyier
790
250k
How STYLIGHT went responsive
nonsquared
98
5.4k
jQuery: Nuts, Bolts and Bling
dougneiner
63
7.6k
Transcript
Neural Relation Extraction for Knowledge Base Enrichment Bayu Distiawan Trisedya,
Gerhard Weikum, Jianzhong Qi, Rui Zhang ACL 2019 հऀ: দాߞ࢙ (ཧݚAIP/౦େ) ୈ11ճ࠷ઌ NLP ษڧձ (γϣʔτ) εϥΠυதͷਤදಛʹऍ͕ͳ͍ݶΓจ͔ΒͷҾ༻Ͱ͢ 1 YAGOͷ։ൃऀ
• MEKG'"$) >9 GD – KG Enrichment
• =1 (: Seq2Seq 2 • @C – ?F#-*>986 IB!0' '34L< – K7IB JA;",+&, n-gram Attention – ?F#-*5).+/*/ % Precision:H 2
: (Wikidata) 3
KG Enrichment Information Extraction ΤϯςΟςΟͱؔΛ WikidataͷIDʹਖ਼نԽ͍ͨ͠ NED => ؔநग़
ͳύΠϓϥΠϯ͕ϕʔεϥΠϯ 4
: New York University
is a private university in Manhattan. Encoder (LSTM) N-gram Attention Decoder (LSTM) Triple Classifier <Q49210 P31 Q902104> <Q49210 P131 Q11299> ᶄ ᶅ Joint Embedding ᶃ ೖྗจ ग़ྗ 5 ೖग़ྗͷϖΞ Distant SupervisionͰੜ
: KG • TransE +
Skip-gram – Anchor Context Model [Yamada 16] 6 New York University is a private university in Manhattan Q49210 is a Q902104 in Q11299 J = +
: N-gram Attention • Entity-mention
– ”New York University” • “New York”, “York University” … 7 3-gram·Ͱͷn-gram
: Triple Classifier • 5<"+(/ 5< (-% –
9: <New York University, capital of, Manhattan> • TransE '- 04&,)-= 7@23$#*!&- 6;.> – (h + r – t) – .>: Negative Sample 81? 8
• ;87/! Distant Supervision – Co-reference – -4#(6#'
3 2 • ;8!* ) ., &!'1 – <New York University, Q49210>$ ;8!9 • ) .,0: "% + (533.5) 9
10 ύΠϓϥΠ ϯμϝ N-gram Attention ޮՌൃش ClassifierͰ Precisionվળ
New York University is a private university in Manhattan.
Encoder (LSTM) N-gram Attention Decoder (LSTM) Triple Classifier <Q49210 P31 Q902104> <Q49210 P131 Q11299> ᶄ ᶅ Joint Embedding ᶃ ೖྗจ ग़ྗ 11 ೖग़ྗͷϖΞ Distant SupervisionͰੜ
• ! • – DS
train/test test 12