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

構文片の改善と評判分析・自動要約への適用

 構文片の改善と評判分析・自動要約への適用

瀧川 和樹 構文片の改善と評判分析・自動要約への適用. 長岡技術科学大学修士論文. (2013.3)

More Decks by 自然言語処理研究室

Other Decks in Research

Transcript

  1. ߏจยͷվળͱ
    ධ൑෼ੳɾࣗಈཁ໿΁ͷ
    ద༻
    ిؾܥɹࢁຊݚڀࣨ

    ୍઒࿨थ

    View full-size slide

  2. 2
    ݚڀഎܠʙॲཧ୯Ґͷ໰୊ʙ
    n  ୯ޠू߹
    ¨ ̍୯ޠͰ͸ҙຯ͕Θ͔Βͳ͍ཁૉ͕ଘࡏ(ޠٛᐆດੑ)
    ex.) ʮ͔͚Δʯ|ʮ෰Λʯʁʮ໎࿭Λʯʁʮ࣌ؒΛʯʁ
    n  ୯ޠn-gram
    ¨ ҙຯͷͳ͍ཁૉ(ෆཁͳσʔλ)͕େྔʹੜ੒͞Εͯ͠·͏
    ex.) ʮ͕,͔͚Δ(2-gram)ʯʮͰ,͋Δ,͜ͱ(3-gram)ʯ
    ݴޠॲཧʹ͓͚Δओͳॲཧ୯Ґ

    View full-size slide

  3. 3
    ݚڀഎܠʙॲཧ୯Ґͷ໰୊ʙ
    n  ୯ޠू߹
    ¨ ̍୯ޠͰ͸ҙຯ͕Θ͔Βͳ͍ཁૉ͕ଘࡏ(ޠٛᐆດੑ)
    ex.) ʮ͔͚Δʯ|ʮ෰Λʯʁʮ໎࿭Λʯʁʮ࣌ؒΛʯʁ
    n  ୯ޠn-gram
    ¨ ҙຯͷͳ͍ཁૉ(ෆཁͳσʔλ)͕େྔʹੜ੒͞Εͯ͠·͏
    ex.) ʮ͕,͔͚Δ(2-gram)ʯʮͰ,͋Δ,͜ͱ(3-gram)ʯ
    ୯ޠͷҙຯΛอ࣋Ͱ͖Δॲཧ୯Ґͷඞཁੑ
    ݴޠॲཧʹ͓͚Δओͳॲཧ୯Ґ

    View full-size slide

  4. 4
    ݚڀഎܠʙॲཧ୯Ґͷ໰୊ʙ
    n  ୯ޠू߹
    ¨ ̍୯ޠͰ͸ҙຯ͕Θ͔Βͳ͍ཁૉ͕ଘࡏ(ޠٛᐆດੑ)
    ex.) ʮ͔͚Δʯ|ʮ෰Λʯʁʮ໎࿭ʯΛʯʁʮ࣌ؒΛʯʁ
    n  ୯ޠn-gram
    ¨ ҙຯͷͳ͍ཁૉ(ෆཁͳσʔλ)͕େྔʹੜ੒͞Εͯ͠·͏
    ex.) ʮ͕,͔͚Δ(2-gram)ʯʮͰ,͋Δ,͜ͱ(3-gram)ʯ
    ୯ޠͷҙຯΛอ࣋Ͱ͖Δॲཧ୯Ґͷඞཁੑ
    ݴޠॲཧʹ͓͚Δओͳॲཧ୯Ґ
    “ߏจยzΛఏҊ

    View full-size slide

  5. 5
    n  ҙຯΛ࣋ͬͨཁૉΛѻ͏͜ͱ͕໨త
    n  ܎Γड͚ͷର͔Βੜ੒
    ݚڀഎܠʙߏจยͱ͸ʙ

    View full-size slide

  6. 6
    n  ҙຯΛ࣋ͬͨཁૉΛѻ͏͜ͱ͕໨త
    n  ܎Γड͚ͷର͔Βੜ੒
    ࠷ۙ·ΘΓͷ૽Ի͕ͱͯ΋େ͖͍
    ݚڀഎܠʙߏจยͱ͸ʙ

    View full-size slide

  7. 7
    n  ҙຯΛ࣋ͬͨཁૉΛѻ͏͜ͱ͕໨త
    n  ܎Γड͚ͷର͔Βੜ੒
    ࠷ۙˠେ͖͍ ·ΘΓͷˠ૽Ի
    ૽Ի͕ˠେ͖͍ ͱͯ΋ˠେ͖͍
    ݚڀഎܠʙߏจยͱ͸ʙ
    ࠷ۙ·ΘΓͷ૽Ի͕ͱͯ΋େ͖͍

    View full-size slide

  8. 8
    n  ҙຯΛ࣋ͬͨཁૉΛѻ͏͜ͱ͕໨త
    n  ܎Γड͚ͷର͔Βੜ੒
    ߏจย
    ݚڀഎܠʙߏจยͱ͸ʙ
    ࠷ۙˠେ͖͍ ·ΘΓͷˠ૽Ի
    ૽Ի͕ˠେ͖͍ ͱͯ΋ˠେ͖͍
    ࠷ۙ·ΘΓͷ૽Ի͕ͱͯ΋େ͖͍

    View full-size slide

  9. 9
    n  ଞͷॲཧ୯Ґͱಉ༷ɺ౷ܭ΋ͱΓ΍͍͢
    n  ܎Γड͚ͷରͰ͋ΔͨΊɺநग़͕༰қ
    ¨ ߏจղੳثΛ༻͍Δ͜ͱͰநग़Մೳ

    ͦͷଞͷॲཧ୯Ґͱൺ΂ͯ
    n  ୯ޠू߹ͱൺ΂ޠٛᐆດੑʹରԠͰ͖Δ
    ¨ ʮ໎࿭-Λˠ͔͚Δʯʮ༸෰-Λˠ͔͚Δʯ
    n  n-gramͱൺ΂จߏ଄͕อ࣋͞Ε͍ͯΔ
    ¨ ෆཁͳσʔλ͕ੜ੒͞Εʹ͍͘
    ݚڀഎܠʙߏจยͷϝϦοτʙ

    View full-size slide

  10. 10
    ࣅͨҙຯͷཁૉ͕ผ෺ͱͯ͠ѻΘΕΔ
    ʢաૄੑͷ໰୊ʣ
    είΞ͕ผʑʹ
    Χ΢ϯτ͞ΕΔ
    Өڹେ
    ౷ܭΛͱΔͱ͖
    ࣙॻͱͯ͠࢖༻͢Δͱ͖
    ݚڀഎܠʙߏจยͷ໰୊(1)ʙ
    ཁૉ ग़ݱස౓
    ૽Ի͕େ͖͍ ̏
    ૽Ի΋େ͖͍ ̐
    େ͖͍૽Ի ̎

    View full-size slide

  11. 11
    ຬ଍͢Δ͜ͱ͕Ͱ͖Δ
    Ұ෦ͰҙຯΛ࣋ͨͳ͍ཁૉ͕ੜ੒͞ΕΔ
    ݚڀഎܠʙߏจยͷ໰୊(2)ʙ
    ຬ଍͢Δˠ͜ͱ
    ͜ͱ-͕ˠͰ͖Δ
    म০ؔ܎ͱ͸ݴ͍ͮΒ͍
    ҙຯ͕௨͡ͳ͍
    ҙຯΛ࣋ͬͨཁૉΛѻ͏ͱ͍͏໨తʹ൓͢Δ

    View full-size slide

  12. 12
    ຬ଍͢Δ͜ͱ͕Ͱ͖Δ
    Ұ෦ͰҙຯΛ࣋ͨͳ͍ཁૉ͕ੜ੒͞ΕΔ
    ݚڀഎܠʙߏจยͷ໰୊(2)ʙ
    ຬ଍͢Δˠ͜ͱ
    ͜ͱ-͕ˠͰ͖Δ
    म০ؔ܎ͱ͸ݴ͍ͮΒ͍
    ҙຯ͕௨͡ͳ͍
    ҙຯΛ࣋ͬͨཁૉΛѻ͏ͱ͍͏໨తʹ൓͢Δ
    ໨తɿ
    ߏจยͷ໰୊఺ͷվળ

    View full-size slide

  13. 13
    n  ߏจยͷ൚Խʢաૄੑͷ໰୊ʣ
    –  ಉྨදݱͷ౷Ұ
    –  ্Ґޠ΁ͷ׵ݴ
    –  ػೳಈࢺͷϥϕϧ෇༩

    n  ܗࣜత಺༰ޠͷ݁߹ʢҙຯΛ΋ͨͳ͍ཁૉͷੜ੒ʣ
    ఏҊख๏

    View full-size slide

  14. 14
    ಉྨදݱͷ౷Ұ
    n  ߏจยͷதʹ͸΄΅ಉ͡ҙຯͷදݱ
    ʹಉྨදݱ͕ଘࡏ
    ಉྨදݱΛϧʔϧʹج͖ͮ൚Խ
    είΞ͕ผʑʹ
    Χ΢ϯτ͞ΕΔ
    ཁૉ ग़ݱස౓
    ૽Ի͕େ͖͍ ̏
    ૽Ի΋େ͖͍ ̐
    େ͖͍૽Ի ̎

    View full-size slide

  15. 15
    ໊ࢺ(-֨ॿࢺ) → ܗ༰ࢺ
    ܗ༰ࢺˠ໊ࢺ
    ཁૉʹؚ·ΕΔ಺༰ޠ͕શͯҰக͍ͯ͠Δ
    ໊ࢺ(-֨ॿࢺ) → ಈࢺ
    ಈࢺɹˠ໊ࢺ
    ࢠڙ-͕ˠتΜͰ͍Δ
    تΜͰ͍Δˠࢠڙ
    ૽Ի-͕ˠ͏Δ͍͞
    ͏Δ͍͞ˠ૽Ի
    or
    (i)
    (ii)
    ಉྨදݱͷ౷Ұ

    View full-size slide

  16. n  γιʔϥεͷ্ҐԼҐ֓೦Λ༻͍ͯɺ୯ޠΛ
    ্ͷ֓೦ʹ׵ݴ͢Δ
    ex.)νϫϫˠݘˠᄡೕྨˠ|

    ্Ґޠ΁ͷ׵ݴ
    ˞γιʔϥε
    ݴ༿Λಉٛޠɺ্ҐɾԼҐ֓೦
    ͳͲͷ؍఺ʹ͓͍ͯ෼ྨͨ͠
    ࣙॻͷ͜ͱ

    View full-size slide

  17. n  ۩ମతͳख๏ɾ৚݅
    ɾγιʔϥεʹ͸EDR֓೦ࣙॻΛ࢖༻
    ʢEDRɾɾɾ໿40ສޠʹ͍ͭͯ෼ྨ͕ͳ͞Εͨγιʔϥεʣ
    ɾߏจย಺ʹ͓͚Δ໊ࢺɾಈࢺΛ্Ґ֓೦ʹ׵ݴ
    ɾ׵ݴ͢Δ֊૚ͷ্ݶ͸બ୒Ͱ͖ΔΑ͏ʹ͢Δ
    ্Ґޠ΁ͷ׵ݴ

    View full-size slide

  18. n  ػೳಈࢺͱ͸
    ໊ࢺʹґଘ͠ɺͦΕࣗ਎ͷҙຯΛ΄ͱΜͲ΋ͨ
    ͳ͍ಈࢺͷ͜ͱ
    ex.)ӨڹΛड͚Δɹ˺ɹӨڹ͞ΕΔ
    n  ͜ͷಛ௃Λར༻ͯ͠ػೳಈࢺΛؚΉ΋ͷʹϥ
    ϕϧΛ෇༩
    n  ϥϕϧʹج͖ͮߏจยΛ൚Խ
    ػೳಈࢺͷϥϕϧ෇༩

    View full-size slide

  19. n  ۩ମతͳख๏
    ɾਓखͰػೳಈࢺΛऩू
    ɾػೳಈࢺΛؚΉߏจยΛ݁߹ͤ͞ɺ
    ̍ͭͷจઅʹ
    ɾػೳಈࢺ͸ͦΕͧΕʮଶʯʮ૬ʯʹ෼ྨ
    ɾ෼ྨ͝ͱʹϥϕϧΛ෇༩
    ػೳಈࢺͷϥϕϧ෇༩
    ˞ଶɿೳಈଶ(౰ͯΔ)ɺडಈଶʢ౰ͯΒΕΔʣ
    ɹ૬ɿਐߦ૬ʢ౰͍ͯͯΔʣɺىಈ૬ʢ౰ͯͩͨ͠

    ͳͲ

    View full-size slide

  20. ڧҙ૬ (ӨڹΛڧΊΔ)
    ؇࿨૬ (ӨڹΛӮΒ͢)
    ػೳಈࢺͷϥϕϧ෇༩ʹ͓͚Δ൚ԽͷྫΛਤ 4.4 ʹࣔ͢ɻ
    ػೳಈࢺͷϥϕϧ෇༩
    “ىಈ૬”ͱ͍͏λάΛ෇༩
    “ىಈ૬”ͱ͍͏λάΛ෇༩

    View full-size slide

  21. 21
    ܗࣜత಺༰ޠͷ݁߹
    ຬ଍͢Δ͜ͱ͕Ͱ͖Δ
    ຬ଍͢Δˠ͜ͱ
    ͜ͱ-͕ˠͰ͖Δ
    म০ؔ܎ͱ͸ݴ͍ͮΒ͍
    ҙຯ͕௨͡ͳ͍

    View full-size slide

  22. 22
    ຬ଍͢Δ͜ͱ͕Ͱ͖Δ
    ຬ଍͢Δˠ͜ͱ
    ͜ͱ-͕ˠͰ͖Δ
    म০ؔ܎ͱ͸ݴ͍ͮΒ͍
    ҙຯ͕௨͡ͳ͍
    ʮ͜ͱʯ͕࣮࣭ػೳతදݱ
    (ʹܗࣜత಺༰ޠ)
    ͱͯ͠ѻΘΕ͍ͯΔ͜ͱ͕໰୊
    ܗࣜత಺༰ޠͷ݁߹

    View full-size slide

  23. 23
    n  ʮ͜ͱʯͷΑ͏ͳػೳతʹѻΘΕΔ୯ޠ(ܗࣜ
    త಺༰ޠ)Λऩू
    n  ͜ΕΒͷ୯ޠ͕͋Δ৔߹ɺ௚લͷ಺༰ޠͷ
    ػೳදݱͱͯ͠ѻ͏
    ຬ଍͢Δ͜ͱ͕Ͱ͖Δ
    ຬ଍͢Δˠ͜ͱ ͜ͱ-͕ˠͰ͖Δ
    ܗࣜత಺༰ޠͷ݁߹

    View full-size slide

  24. 24
    n  ʮ͜ͱʯͷΑ͏ͳػೳతʹѻΘΕΔ୯ޠ(ܗࣜ
    త಺༰ޠ)Λऩू
    n  ͜ΕΒͷ୯ޠ͕͋Δ৔߹ɺ௚લͷ಺༰ޠͷ
    ػೳදݱͱͯ͠ѻ͏
    ຬ଍͢Δ͜ͱ͕Ͱ͖Δ
    ຬ଍͢Δˠ͜ͱ ͜ͱ-͕ˠͰ͖Δ
    ຬ଍͢Δ͜ͱ-͕ˠͰ͖Δ
    ܗࣜత಺༰ޠͷ݁߹

    View full-size slide

  25. 25
    ධ൑෼ੳ΁ͷద༻

    View full-size slide

  26. 26
    ධ൑෼ੳͷ࣮ݧํ๏
    λεΫɿจ෼ྨ
    (1) ධ൑දݱ(ߏจย)rۃੑʢߠఆɾ൱ఆʣείΞɹ
    ͷରΛڭࢣσʔλ͔Βநग़
    (2) (1)ͷ৘ใΛࣙॻʹొ࿥
    (3) ొ࿥ͨࣙ͠ॻͱେن໛ίʔύεΛ༻͍ͯࣙॻΛ
    ֦ு
    (4) ࣙॻΛ༻͍ͯೖྗจͷ֤දݱʹۃੑείΞΛ
    ෇༩
    (5) ۃੑείΞͷ૯࿨͔ΒจΛߠఆ/ ൱ఆʹ෼ྨ

    View full-size slide

  27. 27
    ϑΝϯͷ૽Ի͕େ͖͍
    ϑΝϯ-ͷˠ૽Ի
    ૽Ի-͕ˠେ͖͍
    େ͖͍ˠ૽Ի:൱ఆ
    நग़͞Εͨߏจย
    Ϛονϯά
    ʢಉྨදݱͷ౷Ұʣ
    ϑΝϯ-ͷˠ૽Ի
    ૽Ի-͕ˠେ͖͍ɿ൱ఆ
    ߏจยͷۃੑ
    ೖྗจɿ൱ఆจ
    จ෼ྨ
    ೖྗจ
    ࣙॻ

    View full-size slide

  28. 28
    ϑΝϯͷ૽Ի͕େ͖͍
    ϑΝϯ-ͷˠ૽Ի
    ૽Ի-͕ˠେ͖͍
    େ͖͍ˠ૽Ի:൱ఆ
    நग़͞Εͨߏจย
    Ϛονϯά
    ʢಉྨදݱͷ౷Ұʣ
    ϑΝϯ-ͷˠ૽Ի
    ૽Ի-͕ˠେ͖͍ɿ൱ఆ
    ߏจยͷۃੑ
    ೖྗจɿ൱ఆจ
    ࣙॻ
    จ෼ྨ
    ೖྗจ

    View full-size slide

  29. 29
    ϑΝϯͷ૽Ի͕େ͖͍
    େ͖͍ˠ૽Ի:൱ఆ
    ϑΝϯ-ͷˠ૽Ի
    ૽Ի-͕ˠେ͖͍ɿ൱ఆ
    ߏจยͷۃੑ
    ೖྗจɿ൱ఆจ
    ࣙॻ
    จ෼ྨ
    ೖྗจ
    ϑΝϯ-ͷˠ૽Ի
    ૽Ի-͕ˠେ͖͍
    நग़͞Εͨߏจย
    Ϛονϯά
    ʢಉྨදݱͷ౷Ұʣ

    View full-size slide

  30. ධ൑෼ੳɿ࣮ݧ݁Ռ
    n  ద߹཰͕޲্ͨ͠ख๏
    n  ܗࣜత಺༰ޠͷ݁߹
    n  ࠶ݱ཰͕޲্ͨ͠ख๏
    n  ্Ґޠ΁ͷ׵ݴʢ໊ࢺɾಈࢺͱ΋ʹʣ
    n  ద߹཰ɾ࠶ݱ཰྆ํͱ΋޲্ͨ͠ख๏
    n  ಉྨදݱͷ౷Ұ
    n  ಈࢺͷ্Ґޠ΁ͷ׵ݴ
    n  ͲͪΒ΋޲্͠ͳ͔ͬͨख๏
    n  ػೳಈࢺͷϥϕϧ෇༩

    View full-size slide

  31. ධ൑෼ੳɿߟ࡯
    n  ΄ͱΜͲͷख๏Ͱਫ਼౓͕޲্
    n  ධ൑෼ੳʢ̎஋෼ྨʣʹ͓͍ͯఏҊख๏͕༗ޮ
    n  ൚Խख๏
    →ࣙॻͷొ࿥਺ɺ֦ு਺͕૿Ճ
    n  ܗࣜత಺༰ޠͷ݁߹
    →ҙຯͷͳ͍දݱͷ਺͕ݮগ
    e.g.) ϓϨθϯτʹͳΔ-ͱ→ࢥ͏
    n  ػೳಈࢺͷϥϕϧ෇༩ͷΈਫ਼౓ͷ޲্ͳ͠
    ػೳಈࢺɿݻ͍දݱ͕ଟ͍
    →ϨϏϡʔจͷΑ͏ͳWEB্ͷจষʹ͸ෆ޲͖ʁ

    View full-size slide

  32. 32
    ࣗಈཁ໿΁ͷద༻

    View full-size slide

  33. ࣗಈཁ໿΁ͷద༻
    ৽ฉهࣄ͔ΒॏཁจΛநग़
    நग़͢ΔͨΊͷॏཁ౓ͱͯ͠tf*idfΛ࢖༻
    tf : จॻ಺ʹ͓͚Δ୯ޠͷग़ݱස౓
    idf : ෳ਺จॻ಺ʹ͓͚Δ୯ޠͷग़ݱ͠ʹ͘͞

    View full-size slide

  34. ࣗಈཁ໿ɿtf*idf

    View full-size slide

  35. ࣗಈཁ໿ɿtf*idf
    ॳ೔ͷग़ɿtf͕ߴ͍

    View full-size slide

  36. ࣗಈཁ໿ɿtf*idf
    ॳ೔ͷग़ɿtf͕ߴ͍
    idf͕ߴ͍

    View full-size slide

  37. ࣗಈཁ໿ɿtf*idf
    ॳ೔ͷग़ɿtf͕ߴ͍
    idf͕ߴ͍
    ૣ͘ɿtf͕௿͍

    View full-size slide

  38. ࣗಈཁ໿ɿtf*idf
    ॳ೔ͷग़ɿtf͕ߴ͍
    idf͕ߴ͍
    ૣ͘ɿtf͕௿͍
    idf͕௿͍

    View full-size slide

  39. ࣗಈཁ໿ɿ࣮ݧ݁Ռ
    n  ਫ਼౓͕޲্ͨ͠ख๏
    n  ಉྨදݱͷ౷Ұ
    n  ্Ґޠ΁ͷ׵ݴ(໊ࢺɾಈࢺͱ΋ʹ)
    n  ܗࣜత಺༰ޠͷ݁߹
    n  ਫ਼౓͕޲্͠ͳ͔ͬͨख๏
    n  ػೳಈࢺͷϥϕϧ෇༩

    View full-size slide

  40. ࣗಈཁ໿ɿߟ࡯
    n  ΄ͱΜͲͷख๏Ͱਫ਼౓͕޲্
    n  ࣗಈཁ໿ʹ͓͍ͯ΋ఏҊख๏͕༗ޮ
    n  ػೳಈࢺͷϥϕϧ෇༩ͷΈਫ਼౓͕ѱԽ
    ػೳಈࢺɿग़ݱ͸͢Δ͕ɺ൚Խ͢Δ΄Ͳදݱ͕ࡉ෼
    Խ͞Ε͍ͯͳ͍
    ͦ΋ͦ΋ػೳಈࢺͷ൑ผਫ਼౓͕௿͍

    View full-size slide

  41. 41
    ·ͱΊ
    n  ߏจยͷվྑͷͨΊ̎ͭͷΞϓϩʔνΛ༻ҙ
    n  վྑΛߦͬͨߏจยΛධ൑෼ੳɾࣗಈཁ໿ʹ
    ద༻
    n  ैདྷͷߏจยΑΓ΋શମతʹਫ਼౓͕޲্


    ࠓޙͷ՝୊
    n  ػೳಈࢺͷϥϕϧ෇༩ͷਫ਼౓վળ
    n  ఏҊख๏ͷ૊Έ߹ΘͤʹΑΔਫ਼౓ධՁ

    View full-size slide

  42. ͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝·ͨ͠

    View full-size slide

  43. ධ൑෼ੳͷ݁Ռ

    View full-size slide

  44. 44
    ࣮ݧ݁Ռʢಉྨදݱͷ౷Ұʣ
    ద߹཰ɾ࠶ݱ཰ͱ΋ʹϕʔεϥΠϯΑΓ޲্
    ॲཧ୯Ґ ࠶ݱ཰(%) ద߹཰(%)
    ಉྨදݱͷ౷Ұ 49.8 77.1
    ϕʔεϥΠϯ 48.2 75.5

    View full-size slide

  45. ࣮ݧ݁Ռʢ্Ґޠ΁ͷ׵ݴʣ
    ࠶ݱ཰ɿϕʔεϥΠϯΑΓ޲্
    ద߹཰ɿಈࢺͷஔ׵ͷΈ޲্ େ͖ͳࠩ͸ͳ͍

    ॲཧ୯Ґ ࠶ݱ཰(%) ద߹཰(%)
    ໊ࢺͷ্Ґޠ׵ݴ 54.4 72.6
    ಈࢺͷ্Ґޠ׵ݴ 51.5 76.2
    ໊ࢺɾಈࢺͷ্Ґޠ׵ݴ 59.4 73.6
    ϕʔεϥΠϯ 48.2 75.5
    ൚Խͤ͞Δ໊ࢺɾಈࢺͱ΋ʹ֊૚͸̎֊૚໨·Ͱʹݻఆ

    View full-size slide

  46. 46
    ࣮ݧ݁Ռʢػೳಈࢺͷϥϕϧ෇༩ʣ
    ॲཧ୯Ґ ࠶ݱ཰(%) ద߹཰(%)
    ػೳಈࢺͷϥϕϧ෇༩ 48.2 75.5
    ϕʔεϥΠϯ 48.2 75.5
    ݁Ռʹ͕ࠩͳ͍
    ˠػೳಈࢺΛؚΉςΩετ͕΄ͱΜͲͳ͔ͬͨ

    View full-size slide

  47. 47
    ࣮ݧ݁Ռ(ܗࣜత಺༰ޠͷ݁߹)
    ద߹཰͕޲্͕ͨ͠࠶ݱ཰͕ݮগ
    ॲཧ୯Ґ ࠶ݱ཰(%) ద߹཰(%)
    ܗࣜత಺༰ޠͷ݁߹ 44.6 77.3
    ϕʔεϥΠϯ 47.1 75.5

    View full-size slide

  48. 48
    ॲཧ୯Ґ ࠶ݱ཰(%) ద߹཰(%)
    ୯ޠ2-gram 78.8 79.9
    ୯ޠ3-gram 75.3 78.0
    ܗࣜత಺༰ޠͷ݁߹ 44.6 77.3
    ಈࢺɾ໊ࢺͷ্Ґޠ׵ݴ 59.4 73.6
    ͦͷଞͷॲཧ୯Ґͱͷൺֱ

    View full-size slide

  49. ࣗಈཁ໿ͷ݁Ռ

    View full-size slide

  50. 50
    ࣮ݧ݁Ռʢಉྨදݱͷ౷Ұʣ
    ॲཧ୯Ґ ਫ਼౓(%)
    ಉྨදݱͷ౷Ұ 34.0
    ϕʔεϥΠϯ 32.6

    View full-size slide

  51. ࣮ݧ݁Ռʢ্Ґޠ΁ͷ׵ݴʣ
    ॲཧ୯Ґ ਫ਼౓(%)
    ໊ࢺͷ্Ґޠ׵ݴ 33.0
    ಈࢺͷ্Ґޠ׵ݴ 33.0
    ໊ࢺɾಈࢺͷ্Ґޠ׵ݴ 33.0
    ϕʔεϥΠϯ 32.6
    ໊ࢺ:̔֊૚໨ɺಈࢺɿ̎֊૚໨ʹݻఆ

    View full-size slide

  52. 52
    ࣮ݧ݁Ռʢػೳಈࢺͷϥϕϧ෇༩ʣ
    ॲཧ୯Ґ ਫ਼౓(%)
    ػೳಈࢺͷϥϕϧ෇༩ 31.8
    ϕʔεϥΠϯ 32.6

    View full-size slide

  53. 53
    ࣮ݧ݁Ռ(ܗࣜత಺༰ޠͷ݁߹)
    ॲཧ୯Ґ ਫ਼౓(%)
    ܗࣜత಺༰ޠͷ݁߹ 32.9
    ϕʔεϥΠϯ 32.6

    View full-size slide

  54. 54
    ॲཧ୯Ґ ਫ਼౓(%)
    ୯ޠ2-gram 31.6
    ୯ޠ3-gram 30.0
    ಉྨදݱͷ౷Ұ 34.0
    ैདྷͷߏจย 32.6
    ͦͷଞͷॲཧ୯Ґͱͷൺֱ

    View full-size slide

  55. n  ଞͷॲཧ୯Ґʹ͓͚Δ໰୊఺
    ্Ґͷ֊૚ʹ׵ݴ͗ͯ͢͠ɺදݱຊདྷͷҙຯ
    ͕ࣦΘΕͯ͠·͏
    ex.) νϫϫˠੜ෺ɹΠϧΧˠੜ෺
    n  ߏจย:จ຺৘ใʹΑΓҰఆҎ্ͷ൚Խ͕๷
    ͛Δ
    ex.) νϫϫ͕૸Δɹˠɹ<ੜ෺>͕૸Δ
    ɹ ΠϧΧ͕ӭ͙ɹˠɹ<ੜ෺>͕ӭ͙
    ্Ґޠ΁ͷ׵ݴͷϝϦοτ

    View full-size slide

  56. 56
    ධ൑෼ੳɿධՁ࣮ݧ
    n  ऩूͨ͠σʔλɿߠఆ1,966จɹ൱ఆ1,019จ
    –  ڭࢣσʔλ4/5
    –  ࣮ݧσʔλ1/5
    n  ֦ுࣙॻ༻ͷେن໛ίʔύεɿ໿31.5ສจ
    n  ࣮ݧख๏
    1.  ֤छఏҊख๏
    2.  ϕʔεϥΠϯ:ैདྷͷߏจย

    View full-size slide

  57. 57
    ࣗಈཁ໿ɿධՁ࣮ݧ
    n  idfͷܭࢉɿ೔ຊܦࡁ৽ฉ2000೥ͷهࣄ̍೥෼
    n  ཁ໿཰ʢѹॖ཰ʣɿ25%Ͱݻఆ
    n  ධՁํ๏ɿਓखͷਖ਼ղσʔλͱൺֱ
    - ̍̌̌هࣄ෼ͷཁ໿ΛͦΕͧΕਓखͰ࡞੒
    n  ࣮ݧख๏
    1.  ֤छఏҊख๏
    2.  ϕʔεϥΠϯ:ैདྷͷߏจย

    View full-size slide