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
文献紹介: A Persona-Based Neural Conversation Model
Search
Yumeto Inaoka
February 28, 2018
Science
0
280
文献紹介: A Persona-Based Neural Conversation Model
2018/02/28の文献紹介で発表
Yumeto Inaoka
February 28, 2018
Tweet
Share
More Decks by Yumeto Inaoka
See All by Yumeto Inaoka
文献紹介: Quantity doesn’t buy quality syntax with neural language models
yumeto
1
140
文献紹介: Open Domain Web Keyphrase Extraction Beyond Language Modeling
yumeto
0
180
文献紹介: Self-Supervised_Neural_Machine_Translation
yumeto
0
130
文献紹介: Comparing and Developing Tools to Measure the Readability of Domain-Specific Texts
yumeto
0
130
文献紹介: PAWS: Paraphrase Adversaries from Word Scrambling
yumeto
0
110
文献紹介: Beyond BLEU: Training Neural Machine Translation with Semantic Similarity
yumeto
0
230
文献紹介: EditNTS: An Neural Programmer-Interpreter Model for Sentence Simplification through Explicit Editing
yumeto
0
290
文献紹介: Decomposable Neural Paraphrase Generation
yumeto
0
190
文献紹介: Analyzing the Limitations of Cross-lingual Word Embedding Mappings
yumeto
0
190
Other Decks in Science
See All in Science
Science of Scienceおよび科学計量学に関する研究論文の俯瞰可視化_LT版
hayataka88
0
1k
拡散モデルの原理紹介
brainpadpr
3
5.3k
多次元展開法を用いた 多値バイクラスタリング モデルの提案
kosugitti
0
200
私たちのプロダクトにとってのよいテスト/good test for our products
camel_404
0
210
Design of three-dimensional binary manipulators for pick-and-place task avoiding obstacles (IECON2024)
konakalab
0
110
Mechanistic Interpretability の紹介
sohtakahashi
0
460
Healthcare Innovation through Business Entrepreneurship
clintwinters
0
170
インフラだけではない MLOps の話 @事例でわかるMLOps 機械学習の成果をスケールさせる処方箋 発売記念
icoxfog417
PRO
2
640
はじめてのバックドア基準:あるいは、重回帰分析の偏回帰係数を因果効果の推定値として解釈してよいのか問題
takehikoihayashi
2
1k
事業会社における 機械学習・推薦システム技術の活用事例と必要な能力 / ml-recsys-in-layerx-wantedly-2024
yuya4
3
270
WCS-LA-2024
lcolladotor
0
160
Factorized Diffusion: Perceptual Illusions by Noise Decomposition
tomoaki0705
0
280
Featured
See All Featured
Six Lessons from altMBA
skipperchong
27
3.5k
Docker and Python
trallard
43
3.2k
Art, The Web, and Tiny UX
lynnandtonic
298
20k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
666
120k
4 Signs Your Business is Dying
shpigford
182
21k
Making Projects Easy
brettharned
116
6k
Raft: Consensus for Rubyists
vanstee
137
6.7k
Visualization
eitanlees
146
15k
GraphQLとの向き合い方2022年版
quramy
44
13k
Building Flexible Design Systems
yeseniaperezcruz
328
38k
Facilitating Awesome Meetings
lara
50
6.2k
KATA
mclloyd
29
14k
Transcript
A Persona-Based Neural Conversation Model Jiwei Li, Michel Galley, Chris
Brockett, Georgios Spithourakis, Jianfeng Gao, and Bill Dolan. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 994 - 1003, 2016. จݙհ` Ԭٕज़ՊֶେֶɹࣗવݴޠॲཧݚڀࣨɹҴԬເਓ
"CTUSBDU wऀҰ؏ੑͷΛѻ͏ฦੜϞσϧ wܦྺελΠϧͷΑ͏ͳݸੑΛೖྗʹՃ wQFSQMFYJUZ #-&6ͷ྆ํͰੑೳ্͕ wਓखධՁͰҰ؏ੑʹ͓͍ͯੑೳ্͕ 2
*OUSPEVDUJPO wେྔͷਓؒରਓؒͷରʹΑΔࣗવͳରγεςϜͷ ߏங͕ΛूΊ͍ͯΔ w܇࿅σʔλͷදతͳฦΛฦ͕͋͢Δ ˠͦͷΑ͏ͳฦͷ͕ߴ͘ͳΓ͍ͨ͢Ί wໃ६ͨ͠ฦΛฦ͢͜ͱ͕͋Δ wຊจͰҰ؏ੑͱݸੑͷʹ͍ͭͯऔΓΉ 3
*OUSPEVDUJPO wେྔͷਓؒରਓؒͷରʹΑΔࣗવͳରγεςϜͷ ߏங͕ΛूΊ͍ͯΔ w܇࿅σʔλͷදతͳฦΛฦ͕͋͢Δ ˠͦͷΑ͏ͳฦͷ͕ߴ͘ͳΓ͍ͨ͢Ί wໃ६ͨ͠ฦΛฦ͢͜ͱ͕͋Δ wຊจͰҰ؏ੑͱݸੑͷʹ͍ͭͯऔΓΉ 4
*OUSPEVDUJPO wେྔͷਓؒରਓؒͷରʹΑΔࣗવͳରγεςϜͷ ߏங͕ΛूΊ͍ͯΔ w܇࿅σʔλͷදతͳฦΛฦ͕͋͢Δ ˠͦͷΑ͏ͳฦͷ͕ߴ͘ͳΓ͍ͨ͢Ί wໃ६ͨ͠ฦΛฦ͢͜ͱ͕͋Δ wຊจͰҰ؏ੑͱݸੑͷʹ͍ͭͯऔΓΉ 5
ؔ࿈ݚڀ w3JUUFSΒ ౷ܭతػց༁ͷͱͯ͠औΓΜͩ w4FSCBOΒ ରཤྺͷґଘؔΛิ͢Δ͜ͱΛ తͱͨ͠֊తFODPEFSEFDPEFSϞσϧΛఏҊ w-JΒ
యܕతԠͷׂ߹ΛݮΒͨ͢Ίʹ ࠷େ .-& Ͱͳ͘૬ޓใྔ ..* Λతؔͱ͢Δ TFRTFRγεςϜΛఏҊ 6
ఏҊϞσϧ 7
ఏҊϞσϧ wதؒϢχοτʹ-45.Λ༻͍ͨ3// w࠷ޙͷग़ྗΛ%FDPEFSʹ͢ 8 &ODPEFS
ఏҊϞσϧ wதؒϢχοτʹ-45.Λ༻͍ͨ3// w&ODPEFSͷग़ྗΛ%FDPEFSʹೖྗ w4QFBLFS&NCFEEJOHΛ֤ӅΕͰՃࢉ 9 %FDPEFS
ఏҊϞσϧ w4QFBLFS.PEFM ฦऀͷݸੑͷΈΛߟྀ 4QFBLFS&NCFEEJOHΛೖྗ w4QFBLFS"EESFTTFF.PEFM ฦऀͱฉ͖खͷ྆ํΛߟྀ ԼࣜͰ4QFBLFS&NCFEEJOHΛ߹ 10
%FDPEJOHBOE3FSBOLJOH ɹ.ೖྗจɹ3ฦจɹc3cฦจ ɹW4QFBLFS*%ɹЕ Ѝௐύϥϝʔλ w#FBN4FBSDI࣌ʹ্ࣜͷධՁؔͰ3FSBOLJOHΛߦ͏ wయܕతͰͳ͍͘จ͕༏ઌ͞ΕΔ wɹɹɹɹɹ3͔Β.Λग़ྗ͢ΔTFRTFRΛֶशͯ͠ܭࢉ 11
σʔληοτ w5XJUUFS1FSTPOB%BUBTFU ݄͔Βϲ݄ͷ5XJUUFS'JSF)PTFΛ༻ ظؒʹճҎ্λʔϯͷձΛͨ͠Ϣʔβʹݶఆ ϢʔβʹΑΔ ͷձؚ͕·ΕΔ ಉϢʔβʹΑΔ݄͔Βϲ݄ͷձΛ ͣͭ։ൃ
ݕূ ςετηοτͱͯ͠ઃఆ ฦऀͷ4QFBLFS*%ͷΈ͕ೖ͍ͬͯΔͨΊ4QFBLFS.PEFM ͷΈʹར༻ 12
σʔληοτ w5XJUUFS4PSEPOJ%BUBTFU 4PSEPOJ ैདྷͷ405"ͱͷൺֱͷͨΊʹ༻ ςετηοτͷΈ༻ ͷձσʔλ ͭͷೖྗจʹରͯ͠࠷େݸͷฦ
ˠ5XJUUFS1FSTPOB%BUBTFUͱͷ#-&6ͷൺֱͰ͖ͳ͍ 13
σʔληοτ w5FMFWJTJPO4FSJFT5SBOTDSJQUT%BUB 57γϦʔζl'SJFOETz l5IF#JH#BOH5IFPSZzͷࣈນ ਓͷओཁਓʹΑΔ ͷձ ͏ͪ։ൃ ςετηοτͱͯͦ͠ΕͧΕ ༻ w0QFO4VCUJUMFT
ϊΠζΛؚΉ.ʙ.ͷࣈນσʔληοτ 5FMFWJTJPO4FSJFT5SBOTDSJQUT%BUBͷن͕খ͍ͨ͞Ί ຊσʔληοτͰυϝΠϯదԠΛߦ͏ 14
ֶशͷৄࡉ wMBZFS-45. w IJEEFODFMMTGPSFBDIMBZFS w#BUDITJ[F w-FBSOJOHSBUF w<>ͷҰ༷ͰύϥϝʔλΛॳظԽ w5ISFTIPMEGPSHSBEJFOUDMJQQJOH w7PDBCVMBSZTJ[F
w%SPQPVUSBUF w#FBNTJ[F 15
݁Ռ w5XJUUFS4PSEPOJEBUBTFUʹ͓͚ΔධՁ w.5CBTFMJOF4.5ʹΑΔख๏ wPVSTZTUFN5XJUUFS1FSTPOB%BUBTFUͰֶशͨ͠ͷ wֶशίʔύεͷن %SPQPVUͷ༻ ରϢʔβͷબผ͕ վળͷཧ༝ͱߟ͑ΒΕΔ 16
݁Ռ w5XJUUFS1FSTPOBEBUBTFUʹ͓͚ΔධՁ w.-&ͷ߹ ..*ͷ߹ ͷվળ wఏҊख๏..*ΑΓ.-&ʹΑΓ༗ӹ 17
݁Ռ w57TFSJFTσʔληοτʹ͓͚ΔධՁ w4QFBLFS.PEFM 4QFBLFS"EESFTTFF.PEFMͷ͍ͣΕ #-&6είΞΛ্ͤ͞Δ wఏҊ͢ΔͭͷϞσϧͷؒʹେ͖ͳҧ͍ͳ͍ ˠਓͷύλʔϯ͕ัଊͰ͖ΔఔσʔλαΠζ͕େ͖͘ͳ͍ 18
݁Ռ w5XJUUFS1FSTPOB%BUBTFUͷ։ൃσʔλͱ 57TFSJFTEBUBTFUͰͦΕͧΕQFSQMFYJUZΛൺֱ w5XJUUFSͷํ͕ߴ͘ͳΔͷϊΠζͷͨΊͱߟ͑ΒΕΔ 19
݁Ռ wϥϯμϜʹਓͷ4QFBLFS&NCFEEJOHΛ 4QFBLFS.PEFMʹೖྗ 20
݁Ռ w4QFBLFS"EESFTTFF.PEFM ͷධՁ wฦऀʹහײͰ͋Δ͜ͱ͕ ୯ޠ͔Β͔Δ wlIJNz͔ΒੑผΛਖ਼͘͠ ೝ͍ࣝͯ͠Δ͜ͱ͕͔Δ 21
ਓखධՁ wΫϥυιʔγϯάΛͬͯग़ྗΛධՁ w4QFBLFS*%ຖʹग़ྗ͕Ұ؏͍ͯ͠Δ͔Λ࣮ݧ wϕʔεϥΠϯͱ1FSTPOB.PEFMͷग़ྗΛൺֱͯ͠ ʮҰ؏͍ͯ͠Δʯ ʮҰ؏͍ͯ͠Δʯ ಉఔͰ͋Δ߹ͷείΞΛ͚Δ
wਓͷධՁऀͷείΞΛฏۉ͠ɺͷ࠶ׂΛߦ͏ 22
ਓखධՁ݁Ռ wಉఔͷ߹Λແࢹ͢Δͱɺͷࣄྫʹ͓͍ͯ 1FSTPOB.PEFM͕ʮҰ؏͍ͯ͠ΔʯʮҰ؏͍ͯ͠Δʯ ͱఆ͞Εͨ wʮҰ؏͍ͯ͠ΔʯΛແࢹ͢Δͱɺ1FSTPOB.PEFM͕ ࣄྫͷͰ༏ҐͱͳΓɺϕʔεϥΠϯʹཹ·Δ 23
࣮ࡍͷग़ྗࣄྫ 24
࣮ࡍͷग़ྗࣄྫ 25
݁ w1FSTPOBCBTFEͷԠੜϞσϧΛఏࣔ w#-&6 QFSQMFYJUZ Ұ؏ੑͷਓखධՁʹ͓͍ͯ ܶతͰͳ͍ͷͷϕʔεϥΠϯΛ্ճΔ݁Ռ wฦऀฉ͖खͷਓΛೖྗ͢Δ͜ͱʹϝϦοτ͕͋Δ͜ͱ ͕4QFBLFS"EESFTTFFϞσϧͷ݁ՌͰࣔ͞Εͨ 26