Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Robust Distant Supervision Relation Extraction via Deep Reinforcement Learning

Robust Distant Supervision Relation Extraction via Deep Reinforcement Learning

第10回最先端 NLP 勉強会にて発表した、下記の論文の紹介スライドです。

Qin et al. Robust Distant Supervision Relation Extraction via Deep Reinforcement Learning. ACL 2018.
https://aclanthology.info/papers/P18-1199/p18-1199

https://sites.google.com/view/snlp-jp/home/2018

A0e65af9a6baff8efb7e632212f5eec3?s=128

Mamoru Komachi

August 03, 2018
Tweet

Transcript

  1. Robust Distant Supervision Relation Extraction via Deep Reinforcement Learning Pengda

    Qin, Weiran Xu, William Yang Wang ACL 2018 εϥΠυதͷਤද͸࿦จ͔ΒҾ༻͞Εͨ΋ͷ খொक <komachi@tmu.ac.jp> ୈ10ճ࠷ઌ୺NLPษڧձ@ཧݚ AIP ೔ຊڮΦϑΟε 2018/08/03
  2. ؔ܎நग़͸஌ࣝάϥϑߏஙͷ Ωʔίϯϙʔωϯτ | Barack Obama is married to Michell Obama.

    →spouse 2 Google ͷ஌ࣝάϥϑ
  3. ؔ܎நग़ͷओཁͳ໰୊͸ σʔλεύʔεωε 3 | ஌ࣝϕʔε͔Βؒ઀ڭࢣ͋Γֶशʢdistant supervisionʣʹΑͬͯ஌ࣝ֫ಘ͢Δख๏͕੝Μ | ஌ࣝϕʔεΛݩʹੜίʔύεʹࣗಈͰϥϕϧΛ ͚ͭɺڭࢣ͋ΓֶशͰ෼ྨثΛֶश͠ɺະ஌ͷ ࣄྫΛநग़͢Δ

    →ؒ઀తͳϥϕϧͳͷͰϊΠδʔͳͷ͕໰୊ ʢfalse positive ͕ͨ͘͞Μ͋Δʣ Barack Obama is married to Michell Obama. →spouse
  4. ؒ઀ڭࢣ͋Γֶशʹ͓͚Δ false positive ͷ໰୊ | ਂ૚ֶशొ৔લͷؒ઀ڭࢣ͋Γֶश { ෳ਺ؔ܎Λߟྀ͠ͳ͍ (Minz et

    al., 2009) { ෳ਺ؔ܎Λಉ࣌ʹֶश (Hofmann et al., 2011; Surdeanu et al., 2012) →ΠϯελϯεΛ໌ࣔతʹෛྫʹ༻͍͍ͯͳ͍ | ਂ૚ֶशొ৔ޙͷؒ઀ڭࢣ͋Γֶश { ӅΕ૚Ͱؤ݈ʹֶश͢Δ͜ͱΛૂ͏͕ɺ1Ϋϥε ʹର͠1Πϯελϯε͔͠ਖ਼ྫʹ༻͍ͳ͍ (Zeng et al., 2014; 2015) { ࣄྫʹର͢ΔΞςϯγϣϯΛ༻͍ͯϊΠζʹର Ԡ (Lin et al., 2016; Ji et al., 2017) →ෛྫʢfalse positiveʣΛߟྀ͍ͯ͠ͳ͍ 4
  5. ڧԽֶशͰؒ઀ڭࢣ͋Γֶशͷ ϊΠζ (FP)ΛऔΓআ͘ʢਤ1ʣ 5

  6. ຊݚڀͷ3ߦ·ͱΊ | ؤ݈ͳؒ઀ڭࢣ͋Γؔ܎நग़ͷͨΊͷਂ૚ڧԽ ֶशϑϨʔϜϫʔΫΛఏҊ | ఏҊख๏͸Ϟσϧʹґଘ͠ͳ͍ͷͰɺͲΜͳؔ ܎நग़ख๏ͱ΋૊Έ߹ΘͤΔ͜ͱ͕Մೳ | χϡʔϥϧؔ܎நग़ͷੑೳ޲্Λݕূ 6

  7. ؒ઀ڭࢣ͋ΓֶशͷͨΊͷ ਂ૚ڧԽֶश | ΤʔδΣϯτ͸ɺؒ઀ڭࢣ͋ΓֶशͰΞϊςʔ τ͞Εͨจʹ͍ͭͯɺؔ܎෼ྨͷੑೳʹج͍ͮ ͯɺͦͷจΛ࢒͔͢औΓআ͔͘ΛܾΊΔ 7

  8. ঢ়ଶ: ཤྺΛ Markov Decision Process ͰϞσϧԽ | จ͸୯ޠ෼ࢄදݱͱҐஔ෼ࢄදݱʹม׵ (Zeng et

    al., 2014) | จ෼ࢄදݱ͸ݱࡏͷจϕΫτϧͱ͜Ε·Ͱʹऔ Γআ͔ΕͨจͷฏۉϕΫτϧͷ࿈݁ 8
  9. ߦಈ: ܇࿅ࣄྫʹΠϯελϯε Λ࢒͔͢औΓআ͔͘ | ֤ؔ܎͝ͱʹ1ΤʔδΣϯτ | ͦͷจΛ࢒͔͢औΓআ͔͘Λܾఆ 9

  10. ใु: F஋্͕͕Δ͔Ͳ͏͔ | ؔ܎෼ྨͷ F 1 Λ༻͍ΔʢΫϥεͷෆۉߧ͕͋Δ ϚϧνΫϥεͳͷͰɺਖ਼ղ཰Λ༻͍ͳ͍ʣ !" =

    $(&' " − &' ")') 10
  11. ํࡦ: 2஋෼ྨثͰؔ܎෼ྨ | ୯७ͳ CNN Λ༻͍ͯؔ܎෼ྨثΛߏங (dos Santos et al.,

    2015) 11
  12. ํࡦޯ഑๏ʹΑΔ܇࿅ ʢ࠶ܝਤ1ʣ 12

  13. ํࡦʹجͮ͘ڧԽֶश ϑϨʔϜϫʔΫʢਤ2ʣ 13 | ใुΛܭࢉ͢ΔͨΊʹ!"#$Λ!% "#$ͱ!& "#$ʹɺ '"#$ Λ'% "#$ͱ'&

    "#$ʹ෼ׂ͠ɺͦΕͧΕͷF 1 ΛٻΊΔ | ؔ܎෼ྨث͸ pre-train ͓ͯ͘͠
  14. ؔ܎நग़࣮ݧ | σʔληοτ { New York Times ίʔύεʹ Freebase ͷؔ܎Λ

    λά෇͚ͨ͠σʔλ (Riedel et al., 2010)→52छ ྨͷؔ܎λά { Stanford NE recognizer Ͱ NE λά෇͚ | ࣮ݧઃఆ { ΤʔδΣϯτ͸ CNNɺ୯ޠ෼ࢄදݱ͸ pre-train ͞Εͨ΋ͷΛ࢖༻ { ؔ܎෼ྨث΋ CNNɺ !" #$%ͱ!& #$%͸ͦΕͧΕ2:1ʹ ͳΔΑ͏ʹௐ੔ʢ'" #$%ͱ'& #$% ͸ͦΕͧΕରԠ͢ Δ!" #$%ͱ!& #$% ͷ2ഒʹͳΔΑ͏ʹϥϯμϜαϯϓ Ϧϯάʣ 14
  15. ਂ૚ڧԽֶशʹΑͬͯ ؔ܎நग़ਫ਼౓͕޲্ʢද1ʣ | Originalʢֶश͠ͳ͍ʣͱൺֱͯؒ͠઀ڭࢣ͋ ΓֶशʢpretrainʣͷޮՌ͋Γ | ڧԽֶशʢRLʣͰ͞Βʹੑೳ޲্ 15

  16. ఏҊख๏ʢ+RLʣ͸Ϟσϧ ʹґଘ͠ͳ͍ʢਤ4ʣ | PCNN+ONE: 1จ͚ͩબͿख๏ Zeng et al. (2015) |

    PCNN+ATT: શจʹΞςϯγϣϯ (Lin et al. (2016) 16
  17. ஌ࣝϕʔε (Freebase) ͱ σʔλ (NYT) ͱͷͣΕʢਤ5ʣ | ྔ͕গͳ͍ؔ܎͸ؒҧ͍ͬͯΔ͜ͱ͕ଟ͍ʢؔ ܎ ID

    ͸ද1ͷ֤ߦʹରԠʣ →͜Ε·Ͱͷख๏͸औΓআ͍͍ͯͳ͔ͬͨͷͰɺ ੑೳѱԽʹͭͳ͕͍ͬͯͨ 17
  18. ڧԽֶशʹΑͬͯϊΠζ (FP) ΛऔΓআ͚Δʢද4ʣ | ؒҧͬͨจ຺Λֶश͢ΔͷΛ๷͙͜ͱ͕Ͱ͖Δ 18

  19. ·ͱΊ | ؤ݈ͳؒ઀ڭࢣ͋Γؔ܎நग़ͷͨΊͷਂ૚ڧԽ ֶशϑϨʔϜϫʔΫΛఏҊ { False positive ΛऔΓআ͘͜ͱʹয఺ | ఏҊख๏͸Ϟσϧʹґଘ͠ͳ͍ͷͰɺͲΜͳؔ

    ܎நग़ख๏ͱ΋૊Έ߹ΘͤΔ͜ͱ͕Մೳ | χϡʔϥϧؔ܎நग़ͷੑೳ޲্Λݕূ 19
  20. ॴײ | ڧԽֶशͱ৘ใநग़͸૬ੑ͕͍͍ (Narasimhan et al., 2016) →ʢϒʔτετϥοϓతख๏Ͱ΋͋Γ͕ͪͳʣ ϊΠζΛऔΓআ͘͜ͱ͕Ͱ͖Δ →೚ҙͷख๏ͷલॲཧͱͯ͠΋࢖͑Δ

    | False positive ͷ໰୊͸ؒ઀ڭࢣ͋Γֶशʹͱͬ ͯ΍͸Γ໰୊ Ͱɺϋʔυͳ੍໿ͱͯ͠࢖͏ͷ͸ ةݥ (Nagesh et al., 2014) | ΠϯελϯεΛू߹ͱͯ͠ѻ͏Ξϓϩʔνͱ૊ Έ߹Θ͍ͤͨ | Ϋϥεؒͷؔ܎ΛϞσϧʹೖΕ͍ͨ 20
  21. ࣭ٙԠ౴ᶃ | Q1: ϊΠζΛݮΒ͍ͨ͠ͱ͍͏͜ͱ͕ͩɺ֤Τ ϙοΫͰਖ਼ྫɾෛྫʹݕূʹ࢖͏σʔλ΋ϊΠ ζ͕ೖ͍ͬͯΔͷͰ͸ͳ͍͔ʁ A1: ࣮ࡍʹೖ͍ͬͯΔՄೳੑ͸͋Δ͕ɺֶश͕ ෆ҆ఆʹͳΒͳ͍Α͏ɺ਺ճͷΤϙοΫͷ݁Ռ Λฏۉͨ͠Γ͢ΔςΫχοΫΛ࢖͍ͬͯΔ

    | Q2: ܇࿅ࣄྫ͕มΘΔͨͼʹֶश͠௚͍ͯ͠Δ ͱܭࢉྔ͕େมͰ͸ͳ͍͔ʁ A2: CNN ͰϞσϧΛ͍ܰͯ͘͠Δͷ͸ɺܭࢉྔ తͳ໰୊΋͋Δͷ͔΋͠Εͳ͍ 21
  22. ࣭ٙԠ౴ᶄ | Q3: ؔ܎෼ྨͷϞσϧʹґଘ͠ͳ͍ͱओுͯ͠ ͍Δ͕ɺ෼ྨثͷϞσϧʹ໌Β͔ʹґଘ͢Δͷ Ͱ͸ʁ A3: ෼ྨʹؔͯ͠͸͔֬ʹλεΫʹԠͯ͡࡞Γ ࠐΜͩํ͕Α͍ͱࢥΘΕΔ͕ɺ࿦จͱͯ͠͸؆ ୯ͳϞσϧͰ΋ੑೳ͕Α͘ͳΔ͜ͱΛਪ͍ͨ͠

    ͷͰɺ͋͑ͯφΠʔϒͳϞσϧʹ͍ͯ͠Δ λεΫʹԠͯ͡෼ྨثΛ࡞Ε͹͞ΒʹΑ͘ͳΔ ͱࢥΘΕΔ 22
  23. ࢀߟจݙ σʔληοτ | Riedel et al.. Modeling Relations and Their

    Mentions without Labeled Set. ECML PKDD 2010. ຊݚڀͷϕʔεϥΠϯ | Zeng et al. Distant Supervision for Relation Extraction via Piecewise Convolutional Neural Networks. EMNLP 2015. | Lin et al. Neural Relation Extraction with Selective Attention over Instances. ACL 2016. 23
  24. ࢀߟจݙ χϡʔϥϧҎલͷ৘ใநग़ʢؒ઀ڭࢣ͋Γʣ | Mintz et al. Distant Supervision for Relation

    Extraction without Labeled Data. ACL 2009. | Hoffmann et al. Knowledge-based Weak Supervision for Information Extraction of Overlapping Relations. ACL 2011. | Surdeanu et al. Multi-instance Multi-label Learning for Relation Extraction. EMNLP 2012. | Nagesh et al. Noisy Or-based Model for Relation Extraction using Distant Supervision. EMNLP 2014. 24
  25. ࢀߟจݙ χϡʔϥϧҎޙͷ৘ใநग़ | Zeng et al. Relation Classification via Convolutional

    Deep Neural Network. COLING 2014. | Narasimhan et al. Improving Information Extraction by Acquiring External Evidence with Reinforcement Learning. EMNLP 2016. | Ji et al. Distant Supervision for Relation Extraction with Sentence-level Attention and Entity Descriptions. AAAI 2017. 25