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

How to start studying NLP

kabayan55
September 29, 2018

How to start studying NLP

以下のイベントでの発表資料です。
https://supporterzcolab.com/event/530/

kabayan55

September 29, 2018
Tweet

More Decks by kabayan55

Other Decks in Programming

Transcript

  1. ʲॳ৺ऀ޲͚ʳ
    ɹ͸͡ΊͯΈΑ͏ʂࣗવݴޠॲཧ
    ɹɹࣗવݴޠॲཧͷੈք΁ɺΑ͏ͦ͜
    αϙʔλʔζ$P-BCษڧձ
    ݄

    LBCBZBO

    View full-size slide

  2. LBCBZBO
    େֶɾେֶӃͷݚڀͰࣗવݴޠॲཧΛ
    ༻͍͍ͯͨ
    ๭8FCܥاۀ৽ଔ೥໨
    σʔλαΠΤϯεΤϯδχΞͱͯ͠ɺ
    ػցֶशͷ஌ࣝΛ׆͔ͨ͠ۀ຿Λ
    ߦ͍ͬͯΔʢࣗવݴޠॲཧ΋ʂʣ
    ࣗݾ঺հ

    View full-size slide

  3. झຯͰࣗવݴޠॲཧͰ
    ༡΂Δఔ౓ʹ
    ͳΓ͍ͨͰ͔͢ʁ
    ࣮຿PSݚڀͰ
    ΰϦΰϦ࢖͏Ϩϕϧʹ
    ͳΓ͍ͨͰ͔͢ʁ

    View full-size slide

  4. झຯͰࣗવݴޠॲཧͰ
    ༡΂Δఔ౓ʹ
    ͳΓ͍ͨͰ͔͢ʁ
    ࣮຿PSݚڀͰ
    ΰϦΰϦ࢖͏Ϩϕϧʹ
    ͳΓ͍ͨͰ͔͢ʁ

    View full-size slide

  5. Agenda
    ࣗવݴޠॲཧͰͰ͖Δ͜ͱ
    ࣗવݴޠॲཧͷษڧ๏

    View full-size slide

  6. Agenda
    ࣗવݴޠॲཧͰͰ͖Δ͜ͱ
    ࣗવݴޠॲཧͷษڧ๏

    View full-size slide

  7. ࣗવݴޠΛίϯϐϡʔλͰॲཧ͢Δ
    ࣗવݴޠɿਓ͕ؒ೔ৗతʹ࢖ͬͯΔݴޠ
    ɹɹɹɹɹྫ
    ೔ຊޠɺӳޠ
    ੜ·Εͨͱ͖͔Β਎ۙʹ͋ΔࣗવݴޠΛ
    ίϯϐϡʔλͰॲཧͰ͖Δͬͯ
    ͳΜ͔ͩͦ͢͝͏ʂ
    ʜʜͱ࠷ॳࢲ͸ࢥ͍·ͨ͠
    ࣗવݴޠॲཧͬͯͳʹʁ

    View full-size slide

  8. ܗଶૉղੳ
    ܗଶૉʢ୯ޠʣʹ෼͚ͯ඼ࢺ൑ผ
    $ mecab
    すもももももももものうち
    すもも 名詞,⼀般,*,*,*,*,すもも,スモモ,スモモ
    も   助詞,係助詞,*,*,*,*,も,モ,モ
    もも  名詞,⼀般,*,*,*,*,もも,モモ,モモ
    も   助詞,係助詞,*,*,*,*,も,モ,モ
    もも  名詞,⼀般,*,*,*,*,もも,モモ,モモ
    の   助詞,連体化,*,*,*,*,の,ノ,ノ
    うち名詞,⾮⾃⽴,副詞可能,*,*,*,うち,ウチ,ウチ
    EOS
    ཁૉٕज़

    View full-size slide

  9. ߏจղੳ
    ཁૉٕज़
    Wikipedia より

    View full-size slide

  10. ҙຯղੳ
    ߏจతᐆດੑ͕͋Δͱ͖͸
    ҙຯղੳ͕ඞཁ

    ʮ಄͕੺͍ڕΛ৯΂Δೣʯ
    தଜ໌༟͞Μ !OLNS@BLJ
    ͷ5XJUUFSΑΓ
    ཁૉٕज़

    View full-size slide

  11. จॻ෼ྨ
    จॻΛΧςΰϦ͝ͱʹ෼͚Δ
    ࣗಈཁ໿
    จষΛࣗಈͰཁ໿͢Δ
    ৘ใநग़
    ΩʔϫʔυΛநग़͢Δ
    ྫʣΠϕϯτ৘ใநग़ɺ঎඼৘ใநग़
    ෳ߹ٕज़

    View full-size slide

  12. ෳ߹ٕज़
    ػց຋༁
    ৘ใݕࡧ
    ࣭໰Ԡ౴

    View full-size slide

  13. ෳ߹ٕज़
    ධ൑෼ੳ

    ϨϏϡʔจ
    Positive
    Negative
    ͜ͷέʔΩ͸͍ͪ͝ͷ
    ؁͕͞ࡍཱͬͯඒຯͰͨ͠ɻ
    ·ͨߪೖ͍ͨ͠Ͱ͢ɻ
    ΫϦʔϜ͕؁͗ͨ͢ɻ
    εϙϯδ͕ύαύαͩͬͨɻ

    View full-size slide

  14. Agenda
    ࣗવݴޠॲཧͰͰ͖Δ͜ͱ
    ࣗવݴޠॲཧͷษڧ๏

    View full-size slide

  15. ࢲ͸PythonΛ࢖༻͍ͯ͠·͢
    Python͕ਓؾʂ
    ϝϦοτ
    ! εΫϦϓτݴޠͳͷͰ͙͢ʹ࣮ߦͰ͖Δ
    ! ๛෋ͳϥΠϒϥϦ
    ɹ/VNQZ 4DJQZ /-5, 4DJLJUMFBSO
    ϓϩάϥϛϯάݴޠ͸ʁ

    View full-size slide

  16. ͓͢͢Ίڭࡐ

    View full-size slide

  17. ݴޠॲཧຊϊοΫ
    http://www.cl.ecei.tohoku.ac.jp/nlp100/

    View full-size slide

  18. ݴޠॲཧຊϊοΫ
    ! ౦޻େͷԬ࡚ઌੜ͕࡞੒ͨ͠
    νϡʔτϦΞϧ
    ! Pythonͷ࿅शʹ΋ͳΔ
    ! ݴޠॲཧʹඞཁͳ࣮૷͸͜͜Ͱֶ΂Δ
    ! GitHubʹίʔυΛ্͛ͯΔͻͱ΋ଟ͘ɺ
    ଞͷਓͷίʔυΛࢀߟʹͰ͖ΔͷͰ
    ಠֶ͠΍͍͢

    View full-size slide

  19. ݴޠॲཧຊϊοΫ

    View full-size slide

  20. ݴޠॲཧຊϊοΫ
    GitHubͰ
    “NLP100knock” ͱ
    ݕࡧ͢Δ͚ͩͰ΋ɺ
    86 ϦϙδτϦ
    ݟ͔ͭΔ
    ˞2018೥9݄ݱࡏ

    ར༻ऀͨ͘͞Μ
    ͍·͢

    View full-size slide

  21. /-1ϓϩάϥϛϯάνϡʔτϦΞϧ
    http://phontron.com/teaching.php

    View full-size slide

  22. /-1ϓϩάϥϛϯάνϡʔτϦΞϧ
    http://phontron.com/teaching.php

    View full-size slide

  23. /-1ϓϩάϥϛϯάνϡʔτϦΞϧ
    ! ΧʔωΪʔϝϩϯେֶͷ
    Graham Neubig ઌੜ͕࡞੒ͨ͠
    νϡʔτϦΞϧ
    ! εϥΠυܗࣜ
    ! ֤νϡʔτϦΞϧʹԋश໰୊͕͋Γɺ
    ٖࣅίʔυͱߨٛεϥΠυΛࢀߟʹ
    ࣮૷͢Δͱཧղ͕ਂ·Δ
    ! ਺ࣜΑΓίʔυΛݟͨ΄͏͕
    ཧղ͠΍͍͢ਓʹ͸ಛʹΦεεϝ

    View full-size slide

  24. /-1ϓϩάϥϛϯάνϡʔτϦΞϧ
    ࢿྉɾԋशσʔλ͸
    ͔͜͜Β
    Ұׅμ΢ϯϩʔυʂ
    https://github.com/neubig/nlptutorial

    View full-size slide

  25. ͱΓ͋͑ͣ
    ͜͜·Ͱษڧ͢Ε͹
    ͍͍͔ͳʁ

    View full-size slide

  26. ΋ͬͱ΋ͬͱ
    ษڧͨ͘͠ͳͬͨΒ

    View full-size slide

  27. झຯͰࣗવݴޠॲཧͰ
    ༡΂Δఔ౓ʹ
    ͳΓ͍ͨͰ͔͢ʁ
    ࣮຿PSݚڀͰ
    ΰϦΰϦ࢖͏Ϩϕϧʹ
    ͳΓ͍ͨͰ͔͢ʁ

    View full-size slide

  28. झຯͰࣗવݴޠॲཧͰ
    ༡΂Δఔ౓ʹ
    ͳΓ͍ͨͰ͔͢ʁ
    ࣮຿PSݚڀͰ
    ΰϦΰϦ࢖͏Ϩϕϧʹ
    ͳΓ͍ͨͰ͔͢ʁ

    View full-size slide

  29. ࣗવݴޠॲཧΛಠश͍ͨ͠ਓͷͨΊʹ
    http://cl.sd.tmu.ac.jp/prospective/prerequisite

    View full-size slide

  30. ࣗવݴޠॲཧΛಠश͍ͨ͠ਓͷͨΊʹ
    ट౎େֶ౦ژͷখொઌੜ͕
    ! ਺ֶ
    ! ӳޠ
    ! ϓϩάϥϛϯά
    ! ػցֶश
    ! ࣗવݴޠॲཧ
    ͷษڧͷ࢓ํʹ͍ͭͯ
    ·ͱΊ͍ͯΔϖʔδ

    View full-size slide

  31. ࣗવݴޠॲཧΛಠश͍ͨ͠ਓͷͨΊʹ
    ࠓճ͸ॳ৺ऀ޲͚ͷߨٛͳͷͰ
    ঺հ͚ͩʹͱͲΊ͓͖ͯ·͕͢
    Կͷษڧ͕ඞཁͰ
    Ͳ͏ษڧ͢΂͖͔
    ஸೡʹΘ͔Γ΍͘͢·ͱ·͍ͬͯΔͷͰ
    ੋඇ͝ཡʹͳͬͯ΄͍͠Ͱ͢ʂ

    View full-size slide

  32. &OKPZ
    4UVEZJOH
    /-1

    View full-size slide