Spacyでお手軽NLP / NLP with spacy

2018/06/13のレトリバセミナーのスライドです

Transcript

1. Ͱ͓खܰ/-1 )JSBNBUTV
   !ϨτϦόηϛφʔ
    ը૾
   IUUQTHJUIVCDPNFYQMPTJPOTQB$ZCMPCNBTUFSXFCTJUFBTTFUTJNHMPHPTWH
2. 
    Tsukuba, M2, NLP himkt
3. 5-
   %3 w 1ZUIPOͷࣗવݴޠॲཧϥΠϒϥϦͰ͋Δ4QB$Zͷ঺հ
   w Wͷ͓΋͠Ζػೳʹ͍ͭͯ
   w 4QB$ZͰ೔ຊޠςΩετΛॲཧ͢Δ
   
4. "CPVU
   4QB$Z
    ը૾
   IUUQTTQBDZJP https://spacy.io
5. IUUQTTQBDZJP "CPVU
   4QB$Z “Industrial-Strength NLP” w /POEFTUSVDUJWF
   UPLFOJ[BUJPO
   w /BNFE
   FOUJUZ
   SFDPHOJUJPO
   w 4VQQPSU
   GPS
   
   MBOHVBHFT
   w 
   TUBUJTUJDBM
   NPEFMT
   GPS
   
   MBOHVBHFT
   w ʜFUD
    IUUQTHJUIVCDPNFYQMPTJPOTQB$Z
6. #BTJD
   VTBHF
   PG
   4QB$Z
   
7. .BOZ
   'FBUVSFT
    ը૾
   IUUQTTQBDZJPVTBHFGBDUTpHVSFT
8. 'BTUFTU
   JO
   UIF
   XPSME w %FQFOEFODZ
   QBSTFSͷύϑΥʔϚϯεൺֱ
   w ࿦จ
   IUUQTBDMXFCPSHBOUIPMPHZ111QEG
   w ࿦จʹ͋Δͷ͸41&&%ͷදͰɼ"DDVSBDZ͸ࣗલͰ࡞ΒΕͨ΋ͷʁ
   w $IPJͷϕϯνϚʔΫ࣌ʹ͸TQB$Z
   W͸ະϦϦʔεͳͷͰOB
    ը૾
   IUUQTTQBDZJPVTBHFGBDUTpHVSFT
9. 4QFFE
   DPNQBSJTPO
   XJUI
   PUIFS
   MJCSBSJFT w ଞͷࣗવݴޠॲཧϥΠϒϥϦͱͷ଎౓ൺֱ
   w ࿦จͰ͸ͳ͘࡞ऀ͕ௐࠪͨ͠΋ͷ
   DPOEVDUFE
   JO
   
   w ϦϙδτϦ
   IUUQTHJUIVCDPNFYQMPTJPOTQBDZCFODINBSLT
    ը૾
   IUUQTTQBDZJPVTBHFGBDUTpHVSFT
10. .PEFM
    DPNQBSJTPO w ݴޠʹΑͬͯ͸ෳ਺ͷαΠζͷϞσϧ͕͋Δ
    FO
    GS
    FT
    w 104
    UBHHFS
    
    /&3
    UBHHFS
    
    %FQFOEFODZ
    QBSTFS
    w 8JUIPVU
    BOZ
    QSFQSPDFTTJOH
    EBUBTFU
     ը૾
    IUUQTTQBDZJPVTBHFGBDUTpHVSFT
11. /-5,
    BOE
    4QB$Z w /-5,
    P⒎FST
    TPNF
    PG
    UIF
    TBNF
    GVODUJPOBMJUZ
    BT
    TQB$Z
    w *O
    DPNQBSJTPO
    UP
    TQB$Z
    /-5,
    UBLFT
    B
    NVDI
    NPSF
 CSPBE
    DIVSDI
    BQQSPBDI
    w TQB$Z
    JT
    BMTP
    NVDI
    NPSF
    QFSGPSNBODFGPDVTTFE
    UIBO
    /-5,
    XIFSF
    UIF
    UXP
    MJCSBSJFT
    QSPWJEF
    UIF
    TBNF
    GVODUJPOBMJUZ
    TQB$ZT
    JNQMFNFOUBUJPO
    XJMM
    VTVBMMZ
    CF
    GBTUFS
    BOE
    NPSF
    BDDVSBUF
     Ҿ༻
    IUUQTTQBDZJPVTBHFGBDUTpHVSFT
12. 0UIFS
    MJCSBSJFT
    BOE
    TQB$Z
     1ZUPSDI
    IUUQTHJUIVCDPNQZUPSDIQZUPSDICMPCNBTUFSEPDTTPVSDF@TUBUJDJNHQZUPSDIMPHPEBSLTWH
    "MMFO/-1
    IUUQTHJUIVCDPNBMMFOBJBMMFOOMQCMPCNBTUFSEPDTUBUJDBMMFOOMQMPHPEBSLQOH
    (FOTJN
    IUUQTHJUIVCDPN3B3F5FDIOPMPHJFTHFOTJNCMPCEFWFMPQEPDTTSDSFBENF@JNBHFTSBSFQOH
    $V1Z
    IUUQTHJUIVCDPNDVQZDVQZCMPCNBTUFSEPDTJNBHFDVQZ@MPHP@QYQOH JO
    UPSDIUFYU GPS
    (16
    BDDFMFSBUJPO XPSEWFDUPS QJQFMJOF
13. 4QB$Z
    Wͷݸਓతʹ޷͖ͳػೳ w EJTQMB$Z
    Ͱ͌͢΀Εʔ͠ʔʁ
    w 4QB$ZͷՄࢹԽϞδϡʔϧ
    w ͖Ε͍Ͱ͍͍ײ͡ͳը૾Λ࡞ͬͯ͘ΕΔ
    w ݻ༗දݱநग़ͱ܎Γड͚ղੳͷ݁ՌΛՄࢹԽͯ͘͠ΕΔ
    w 4QB$ZͰղੳͨ͠ΦϒδΣΫτΛͦͷ··࢖͑Δ
    w 47(ܗࣜͷը૾͕ग़ྗ͞ΕΔ
    
14. 8FCαΠτ্ͷ/&3ͷσϞ
     ը૾
    IUUQTFYQMPTJPOBJEFNPTEJTQMBDZFOU
15. 4QB$Z
    Y
    +VQZUFS
    /PUFCPPL
    
16. 4QB$Z
    BOE
    ೔ຊޠ w WͰͷ೔ຊޠରԠ
    13
    
    
    w ೔ຊޠܗଶૉղੳث+BOPNFΛϥοϓ͢ΔܗͰ࣮૷
    w WͰͷܗଶૉղੳثҠߦ
    *TTVF
    
    13
    
    
    w ೔ຊޠ6OJWFSTBM
    %FQFOEFODZσʔλ͸6OJ%JDͰ෼ׂ͞Ε͍ͯΔ
    w +BOPNF͸ݱࡏͷͱ͜Ζ6OJ%JDʹະରԠ
    w 
    .F$BCʹҠߦ
    
17. 4QB$Z
    BOE
    ೔ຊޠ
    
18. 4QB$Z
    BOE
    ೔ຊޠ
     "OTXFS
    VTF
    6OJ%JD
19. 4QB$Z
    BOE
    ೔ຊޠ
    
20. ೔ຊޠ
    /&3
    
    %FQFOEFODZ
    QBSTJOH
    XJUI
    4QB$Z w ݁࿦
    ݱࡏ͸Ͱ͖ͳ͍
    w 4QB$Zʹ͸ࣗ෼Ͱ5BHHFS΍1BSTFSΛֶशͰ͖Δ
 ػߏ͕૊Έࠐ·Ε͍ͯΔ
    w ೔ຊޠ6OJWFSTBM
    %FQFOEFODJFT͸ެ։͞Ε͍ͯΔ
 IUUQTHJUIVCDPN6OJWFSTBM%FQFOEFODJFT6%@+BQBOFTF(4%
    ͳͲ
     
    Ϟσϧࣗ࡞Ͱ͖ΔͷͰ͸ʂʁ
21. "EEJOH
    -BOHVBHF
    TVQQPSU
     4QB$ZͷݴޠϞδϡʔϧͷ
 ίϯϙʔωϯτ ը૾
    IUUQTTQBDZJPVTBHFBEEJOHMBOHVBHFTTFDUJPOUSBJOJOH
22. "EEJOH
    -BOHVBHF
    TVQQPSU
     ը૾
    IUUQTTQBDZJPVTBHFBEEJOHMBOHVBHFTTFDUJPOUSBJOJOH 1PXFSFE
    CZ
    .F$BC
 
    4QB$ZͷֶशϞδϡʔϧ͕࢖͑ͳ͍ʁ
23. ೔ຊޠTQB$Zͷݱࡏͷ໰୊ʁ w ࣙॻ͕6OJ%JDͰ͋Δ͜ͱ͕૝ఆ͞Ε͍ͯΔ
    JTTVF
    
    w ʢ͓ͦΒ͘ʣଟ͘ͷ؀ڥͰ͸.F$BCͷσϑΥϧτࣙॻ͸*1"EJD
    w <50%0>
    +BQBOFTF5PLFOJ[FSͰ5BHHFSΛ࡞͍ͬͯΔ͕ɼ
 ͜ͷίϯετϥΫλͷҾ਺͸֎෦͔Β৮Εͳͦ͏
    w ൃԻ͕ະొ࿥ͷ୯ޠΛղੳ͢ΔͱΤϥʔ
    *1"EJD
    13
    
    w ࣙॻ͝ͱͷग़ྗͷࠩҟΛٵऩ͢ΔΠϯλʔϑΣʔε͕ඞཁʁ
    w /&3΍%FQFOEFODZ
    1BSTJOHͷϞσϧ͸·ͩଘࡏ͠ͳ͍
    w ݱࡏ͸
    A"MQIB
    UPLFOJ[BUJPO
    TVQQPSUA
    w ෼͔ͪॻ͖͕ඞཁͳݴޠͷରԠ͸Ͳ͏͢Ε͹ʜ
    w தࠃޠ΋ࣅͨঢ়گͰࢭ·͍ͬͯΔ
    13
    
    
24. ͜ΕΛ΍Ε͹ͱΓ͋͑ͣςετ͸Ͱ͖ͦ͏  USBWJTͰ.F$BCΛΠϯετʔϧ͢ΔΑ͏ʹ͢Δ
  TQBDZNPEFMTʹ6OJ%JDΛొ࿥͢Δ
  QZUIPO
    N
    TQBDZ
    EPXOMPBE
    KBΛ੔උ͢Δ
     /&3ͱ%FQFOEFODZ
    1BSTJOH͸͋ͱͰௐ΂Δʜ
25. ·ͱΊ w ࣗવݴޠॲཧϥΠϒϥϦTQB$Zͷ঺հ
    w *OEVTUSZൃͷࣗવݴޠॲཧϥΠϒϥϦ
    w WͰೖͬͨՄࢹԽϞδϡʔϧ͕͍͍ײ͡
    w EJTQMB$ZΛ࢖͍͍ͬͨײ͡ͳՄࢹԽ
    w TQB$ZͰͷ೔ຊޠςΩετॲཧ͸·ͩෆ׬શ
    w ݱঢ়
    ܗଶૉղੳͷΠϯλʔϑΣʔε
     QJQ
    JOTUBMM
    TQBDZ 3VO
    5IJT
    DPNNBOE