$30 off During Our Annual Pro Sale. View Details »

【YANS2022 チュートリアル】学術情報検索と推薦

Marie Katsurai
PRO
September 11, 2022

【YANS2022 チュートリアル】学術情報検索と推薦

NLP若手の会 (YANS) 第17回シンポジウム (2022年8月29日)のチュートリアル講演資料です。
雑談などのスライドは省いてあります。

Marie Katsurai
PRO

September 11, 2022
Tweet

Other Decks in Research

Transcript

  1. ֶज़৘ใݕࡧͱਪન
    ܡҪ ຑཬҥ
    ಉࢤࣾେֶ ཧ޻ֶ෦ ΠϯςϦδΣϯτ৘ใ޻ֶՊ
    :"/4

    View Slide

  2. ߨԋऀ঺հ

    ܡҪ ຑཬҥ
    ೥݄ ๺ւಓେֶ޻ֶ෦৘ใΤϨΫτϩχΫεֶՊଔۀ
    ೥݄ ๺ւಓେֶେֶӃ৘ใՊֶݚڀՊϝσΟΞωοτϫʔΫઐ
    ߈म࢜՝ఔमྃ
    ೥݄ ೔ຊֶज़ৼڵձಛผݚڀһ %$

    ೥݄ ๺ւಓେֶେֶӃ৘ใՊֶݚڀՊϝσΟΞωοτϫʔΫઐ
    ߈ത࢜ޙظ՝ఔमྃ
    ത࢜ʢ৘ใՊֶʣ
    ೥݄ ೔ຊֶज़ৼڵձಛผݚڀһ 1%

    ೥݄ ಉࢤࣾେֶཧ޻ֶ෦৘ใγεςϜσβΠϯֶՊॿڭ
    ೥݄ ಉࢤࣾେֶཧ޻ֶ෦ΠϯςϦδΣϯτ৘ใ޻ֶՊॿڭ
    ೥݄ ʏ ।ڭत
    ΧλΧφֶՊ͹͔ΓͷܦྺͰ͢ʜ

    View Slide

  3. ֶज़ݚڀΛऔΓר͘ࣄ৘

    ిࢠδϟʔφϧͷ੒௕
    ͋ΔௐࠪʹΑΔͱίϯϐϡʔλαΠΤϯεܥͷݚڀऀ͕ಡΜͰ͍Δ࿦จ͸
    ฏۉ ݄ຊʢʹ೥ຊʣʂ <5FOPQJS >
    ⻘線:英語論⽂誌の数
    ⾚線:増加率
    ൈਮݩʀ3+PIOTPO "8BULJOTPO BOE.8BCF l5IF45.SFQPSU"OPWFSWJFXPGTDJFOUJGJDBOETDIPMBSMZ
    QVCMJTIJOH zUIFEJUJPO 0DU
    $5FOPQJS -$ISJTUJBO +,BVGNBO l4FFLJOH 3FBEJOH BOE6TFPG4DIPMBSMZ"SUJDMFT"O*OUFSOBUJPOBM4UVEZPG1FSDFQUJPOTBOE#FIBWJPSPG
    3FTFBSDIFST z 1VCMJDBUJPOT WPM OP

    View Slide

  4. ݚڀऀΛࢧ͑ΔݕࡧΤϯδϯ

    IUUQTTDIPMBSHPPHMFDPKQ

    View Slide

  5. (PPHMF4DIPMBSͷݕࡧΞϧΰϦζϜ
    • ʮ࿦จͷϑϧςΩετɺग़൛৔ॴɺஶऀɺඃҾ༻ճ਺ɺඃҾ༻ͷ઱౓ΛߟྀʯCZ(PPHMF

    IUUQTTDIPMBSHPPHMFDPNJOUMKBTDIPMBSBCPVUIUNM

    View Slide

  6. (PPHMF4DIPMBSͷݕࡧΞϧΰϦζϜ
    • ʮ࿦จͷϑϧςΩετɺग़൛৔ॴɺஶऀɺඃҾ༻ճ਺ɺඃҾ༻ͷ઱౓ΛߟྀʯCZ(PPHMF

    ༗໊ͳஶऀʁ
    ͍ͭͷҾ༻͕ଟ͍ʁ
    ༗໊ͳδϟʔφϧʁ
    ඃҾ༻਺͸ଟ͍ʁ

    View Slide

  7. ஶऀͷධՁࢦඪ
    • ࠷΋؆ศͳࢦඪ͸ʮ࿦จ਺ʯɻ

    ࿦จ਺



    • ࣭Λߟྀ͍ͯ͠ͳ͍ͱ͍͏໰୊͕͋Δɻ

    View Slide

  8. ඃҾ༻
    • ଞऀ͔Βࢀߟʹ͞ΕΔ࿦จ͸࣭͕ߴ͍ͱԾఆɻ

    ࢀߟจݙ

    View Slide

  9. ஶऀͷධՁࢦඪ IJOEFY<)JSTDI 1/"4 >
    • ࿦จͷੜ࢈ੑͱӨڹྗͷ྆ํΛଌఆ͠Α͏ͱ͢Δࢦඪɻ
    ࿦จΠϯσοΫε ඃҾ༻਺







    ඃҾ༻਺ɿ
    IJOEFY
    JJOEFY
    (被引⽤数10以上
    の論⽂数)

    "͞Μ
    +&)JSTDI "OJOEFYUPRVBOUJGZBOJOEJWJEVBMTTDJFOUJGJDSFTFBSDIPVUQVU z 1SPDFFEJOHTPGUIF/BUJPOBM"DBEFNZPG4DJFODF WPM OP
    QQr
    • ௚ۙ೥ͷؒʹग़൛͞Εͨ࿦จʹ͍ͭͯࢉग़ͨ͠৔߹ɿIJOEFY
    • Լهͷར఺͔Βٸ଎ʹ஫໨ΛूΊͨ ɿ
    • Ҿ༻ͷͳ͍࿦จ΍ฏۉҎ্ͷҾ༻਺Λ΋ͭ࿦จͳͲɺۃ୺ͳ஋ʹӨڹ͞Εͳ͍ɻ
    • 8FCPG4DJFODFͳͲͷσʔλϕʔεʹऩ࿥͞Ε͍ͯΔඃҾ༻ճ਺Λιʔτ͢Δ͚ͩͰ؆୯ʹࢉग़Ͱ͖Δɻ
    • ୯Ұͷ஋ͷൺֱͰࡁΉɻ

    View Slide

  10. IJOEFYͷ೿ੜɺHJOEFY<&HHIF *44*/FXTMFUUFS >
    l ඃҾ༻਺ॱͰ࿦จΛฒ΂ͨͱ͖ɼ্Ґ(ݸͷ࿦จ͕গͳ͘ͱ΋(ճͷҾ༻Λड͚ΔΑ͏ͳʢ།Ұ
    ͷʣ࠷େͷ਺ɻ
    • HJOEFY͸্Ґͷ࿦จ͕߹ΘͤͯͷඃҾ༻਺Λಘͨ͜ͱΛࣔ͢ɻ
    • ඃҾ༻਺͕ඇৗʹଟ͍࿦จΛߴ͘ධՁ͢Δ఺͕IJOEFYͱͷࠩҟɻ
    • શͯͷൃද࿦จͷฏۉඃҾ༻਺͕ൃද࿦จͷ૯਺Λ௒͑ͨ৔߹ʹ͏·͘ධՁͰ͖ͳ͍ͱ͍
    ͏ܽ఺Λ΋ͭɻ
    l ଞʹ΋IJOEFYͷ೿ੜ͕਺ଟ͘ఏҊ͞Ε͍ͯΔ͕ɺ͍ͣΕ΋ࢉग़͕ෳࡶͰ͋ͬͨΓܽ఺Λ๊͍͑ͯ
    ΔͨΊɺIJOEFYɺHJOEFYɺ૯ඃҾ༻਺ͷछྨ͕༻͍ΒΕΔ͜ͱ͕ଟ͍ɻ
    ࿦จΠϯσοΫε ඃҾ༻਺







    ඃҾ༻਺ɿ
    IJOEFY
    HJOEFY

    "͞Μ

    View Slide

  11. IJOEFYʹର͢Δ਺ʑͷࢦఠ
    • IJOEFYͷߴ͍ݚڀऀ͸ɼଟ͘ͷڞஶऀʢ૯਺ਓҎ্ʣΛ΋ͪɺޓ͍ʹҾ༻͋͠͏͜ͱ͕ଟ͍
    <.BTJD#FHJD >ɻ
    • ڞஶ࿦จ͕ଟ͍΄Ͳ࿦จ਺ɺඃҾ༻਺͕૿Ճ͢Δ͜ͱ͸Α͘஌ΒΕ͍ͯΔɻ
    • ஶऀͷ͸IJOEFY͕ʙͰ͋Γɺ΄ͱΜͲͷஶऀ͸ޓ͍ʹ۠ผΛ͚ͭΔ͜ͱ͕೉͍͠
    <$SFTQP4JNPFT 442 >ɻ
    • IJOEFYͷࢉग़ର৅Ͱ͋ΔI಺ʹೖΒͳ͍࿦จΛධՁͰ͖ͳ͍ɻ

    ڞஶऀͷྠ
    ඞͣ͠΋ສೳͳࢦඪͰ͸ͳ͍͜ͱʹ஫ҙ͢Δඞཁ͕͋Δɻ

    View Slide

  12. ग़൛৔ॴʢδϟʔφϧɺࠃࡍձٞͳͲʣͷධՁࢦඪ
    • ஶऀͷͱ͖ͱಉ༷ʹIJOEFYܥͷࢦඪ͕ࢉग़͞ΕΔɻ

    δϟʔφϧ
    IF(y) =
    ʢy-1೥ͷܝࡌ࿦จ਺ʣʴʢy-2೥ͷܝࡌ࿦จ਺ʣ
    y-1೥ɺy-2೥ͷܝࡌ࿦จ͕Z೥ʹҾ༻͞Εͨճ਺
    ௚ۙ೥ؒͷඃҾ༻਺͕ଟ͍ॱʹ
    ࿦จΛιʔτ

    IJOEFY
    • *NQBDUGBDUPS *'
    ͱ͍͏ࢦඪ΋༗໊ɻ͜Ε͸ग़൛৔ॴಛ༗ͷࢦඪɻ

    View Slide

  13. Ҿ༻จԽͷҧ͍ʹ΋஫ҙ͢Δඞཁ͕͋Δ
    • Ҿ༻ͷ͞Εํ͸෼໺ʹΑͬͯେ͖͘ҟͳΔͨΊɺಉҰ෼໺಺ͰͷධՁʹ͔͠࢖͑ͳ͍ɻ
    • ೥݄ݱࡏɺ֤෼໺ͷIJOEFYτοϓࡶࢽ͸ҎԼͷͱ͓Γɻ˞(PPHMF4DIPMBS.FUSJDTΑΓ
    • ίϯϐϡʔλϏδϣϯɿ$713
    *$$7
    &$$7

    • ܭࢉݴޠֶɿ"$-
    &./-1
    )-5/""$-

    • ਤॻؗ৘ใֶɿ4DJFOUPNFUSJDT
    +PVSOBMPGUIF"TTPDJBUJPOGPS*OGPSNBUJPO4DJFODFBOE
    5FDIOPMPHZ
    +PVSOBMPG*OGPSNFUSJDT

    • Ҩ఻ֶɾήϊϛΫεɿ/BUVSF(FOFUJDT
    (FOPNF#JPMPHZ

    • ݐஙɿ#VJMEJOHT
    -JHIUJOH3FTFBSDI5FDIOPMPHZ

    • Ҿ༻཰΋෼໺ʹΑͬͯҟͳΔɻ
    • ࡉ๔ੜ෺ֶͷ্Ґࡶࢽͷ*'͸໿ɻ
    • ਺ֶͷ্Ґࡶࢽͷ*'͸໿ɻ

    View Slide

  14. (PPHMF4DIPMBSͷΞϧΰϦζϜௐࠪ <#FFM +41 >
    • ༷ʑͳݕࡧΫΤϦΛ༻͍ͯ࿦จݕࡧ݁ՌΛऩू͠ɺඃҾ༻ճ਺ɾޠͷग़
    ݱස౓ɾग़൛೔ͱදࣔॱҐͷؔ܎ΛϓϩοτɻҎԼͷ͜ͱ͕ࣔࠦ͞Εͨɻ
    ü ඃҾ༻ճ਺ͷଟ͞͸ॱҐʹେ͖͘Өڹ͢Δɻ
    ü ΫΤϦͷޠ͕۟ग़ݱ͢Δճ਺͸Өڹ͠ͳ͍ɻͨͩ͠ɺλΠτϧʹೖ͍ͬͯΔ
    ํ্͕Ґʹ͘Δɻ
    ü ݕࡧΫΤϦͷಉٛޠ͸ߟྀ͞Ε͍ͯͳ͍ɻ
    ü ΫΤϦޠ۟Λஶऀ໊ʹ΋ͭͱ্Ґʹ͘Δɻ

    +#FFM #(JQQ BOE&8JMEF l"DBEFNJD4FBSDI&OHJOF0QUJNJ[BUJPO "4&0
    0QUJNJ[JOH4DIPMBSMZ-JUFSBUVSFGPS(PPHMF4DIPMBSBOE$P z+PVSOBM
    PG4DIPMBSMZ1VCMJTIJOH WPM OP QQ +BOVBSZ
    8IJUF-&%
    ˠ8IJUF͞Μ͕ݕࡧ݁Ռʹʜ

    Ҿ

    ε
    ί
    Ξ
    ݕࡧ݁ՌͷॱҐ
    ಉٛޠΛొ৔
    ͤͨ͞ํ͕
    ݕࡧʹ༗རʜ
    λΠτϧ͸
    ॏཁʜ

    View Slide

  15. (PPHMF4DIPMBSͷϦόʔεΤϯδχΞϦϯά <3PWJSB 1%-* >
    • (PPHMF4DIPMBSͷ֤छࢦఆݕࡧʹΑΔॱҐ͸ඃҾ༻਺͚ͩͰ࠶ݱͰ͖Δ͔Λௐࠪɻ
    • Ͳͷݕࡧํ๏ʹ͓͍ͯ΋ඃҾ༻਺ʹΑΔϥϯΩϯά݁ՌͱͷεϐΞϚϯॱҐ૬ؔ܎਺͕Ҏ্ͱ
    ͳͬͨɻ


    ໊ࣈͰݕࡧ

    ग़൛৔ॴ໊Ͱݕࡧ

    ೥Ͱݕࡧ
    ඃҾ༻਺ʹΑΔ
    ϥϯΩϯά
    $3PWJSB '(VFSSFSP4PMÉ -$PEJOB l3FDFJWFEDJUBUJPOTBTBNBJO4&0 GBDUPSPGHPPHMFTDIPMBSSFTVMUTSBOLJOH z
    1SPGFTJPOBMEF-B*OGPSNBDJPO WPM OP QQr
    ඃҾ༻਺͕ଟ͍࿦จ͕ࠓޙ΋஫໨ΛूΊΔͱ͍͏
    ϚγϡʔޮՌΛ૿େͤ͞Δ࢓૊Έʁ

    View Slide

  16. ༨ஊɿ(PPHMF4DIPMBS͔Βͷσʔλऩू
    • σʔλऩू༻"1*͸ఏڙ͞Ε͍ͯͳ͍ʢʂʣ
    • εΫϨΠϐϯάΛ܁Γฦ͢ʹ͸$"15$)"Λղ͘ඞཁ͕͋Δɻ
    • (PPHMF4DIPMBS্ͷສຊ෼ͷ࿦จσʔλΛ෼ੳͨ͠νʔϜ͸ɺϲ݄͔͚ͯखಈͰ
    $"15$)"Λղ͖ଓ͚ͨͱͷ͜ͱ <&MTF>ɻ
    )&MTF l)PX*TDSBQFEEBUBGSPN(PPHMF4DIPMBS z/BUVSF/&842"
    IUUQTXXXOBUVSFDPNBSUJDMFTE

    ϙνʜ
    ϙνʜ

    View Slide

  17. ࿦จݕࡧͱ࿦จਪન
    • ࿦จΛݕࡧ͢Δʹ͸ॴ๬ͷ಺༰ΛΫΤϦͱͯ͠໌֬ʹೖྗ͢Δඞཁ͕͋Δɻ
    • Ұํɺؔ࿈͢Δ࿦จΛޮ཰తʹ஌Γ͍ͨͱ͍͏χʔζʹ౴͑Δͷ͕࿦จਪનͰ͋Δɻ

    (MPCBMͳ࿦จਪન MPDBMͳ࿦จਪન
    ೳಈతʹݕࡧ͢Δͷ͕
    ͭΒ͍ͷʜ
    ؔ࿈͢Δ࿦จΛ
    ͓͢͢Ί͢ΔΑʂ
    ࢲʹͽͬͨΓͷ
    ࿦จͩΘʂʂʂ
    จষʹର͠Ҿ༻͢΂͖จݙΛਪન
    l-PDBUJOHTDJFOUJGJDTQFDJGJDDPOUFOU
    GSPNBMBSHFDPSQPSBJTDSVDJBMUP
    SFTFBSDIFST< >z
    ࿦จͷओͳ಺༰ʹ஫໨
    ຊߨԋͰ͸ͪ͜ΒΛ঺հ

    View Slide

  18. ࿦จਪનख๏ͷΧςΰϥΠζ
    ڠௐϑΟϧλϦϯάख๏ ίϯςϯτϕʔεख๏
    άϥϑຒΊࠐΈख๏ ϋΠϒϦου

    Ϣʔβͷڵຯ ࿦จͷ಺༰
    4JNJMBS
    4JNJMBS
    Ҿ༻ Ҿ༻

    View Slide

  19. ڠௐϑΟϧλϦϯά DPMMBCPSBUJWFGJMUFSJOH$'

    • ଞϢʔβͷධՁ৘ใΛ༻͍Δਪનख๏ɻ
    • &ίϚʔεͰͷਪનͱ͸ҟͳΓɺ࿦จਪનͰ͸ϢʔβΛߟ͑ͣࢀߟจݙϦετΛ༻͍Δ͜ͱ͕ଟ͍ɻ
    Ø ࿦จʹϨʔςΟϯάΛ༩͑Δͷ͸໘౗ͳͨΊ <:BOH+$%->ɻ

    映画1 映画2 映画3 映画4 …
    Ϣʔβ
    5 3 2
    Ϣʔβ
    3 2 3
    Ϣʔβ
    5 2 3
    ʜ
    論⽂1 論⽂2 論⽂3 論⽂4 …
    ࿦จ
    0 1 0 0
    ࿦จ
    0 0 0 0
    ࿦จ
    0 0 0 1
    ʜ
    ঎඼΍өըͷਪન ߦ͕Ϣʔβɺྻ͕ΞΠςϜͷ
    ϨʔςΟϯάߦྻ
    ࿦จਪન ߦ͕Ҿ༻ݩ࿦จɺྻ͕Ҿ༻ઌ࿦จͷ
    ೋ஋ߦྻ

    View Slide

  20. ڠௐϑΟϧλϦϯάʹΑΔ࿦จਪન <.D/FF$4$8`>
    • ߦͱྻΛ࿦จͱͯ͠ߏஙͨ͠Ҿ༻ߦྻʹ͓͍ͯ L/FBSFTU/FJHICPS L//
    $'Λ࣮ߦɻ
    • ߦϕΫτϧؒͷίαΠϯྨࣅ౓Λࢉग़͠ɺΫΤϦ࿦จͱͷྨࣅ౓্͕Ґ L ݅ͱͳΔ࿦จΛબग़ɻ
    • ֤ۙ๣ͷҾ༻ঢ়گΛྨࣅ౓ͰॏΈ෇͚ͯ͠౤ථɻ

    … p50
    p51
    p52
    p53
    p54 …
    p1

    p2

    p3

    p4

    p5








    ΫΤϦ
    ࿦จ
    TJNJMBS
    Lۙ๣
    Lۙ๣͔Βͷ౤ථ ֤ۙ๣ͱͷྨࣅ౓
    4..D/FF FUBM l0OUIFSFDPNNFOEJOHPGDJUBUJPOTGPSSFTFBSDIQBQFST zJO1SPD$4$8
    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
    Pa,i = ¯
    ra +
    P
    u2N
    (ru,i ¯
    ru) ⇥ sim(a, u)
    P
    u2N
    sim(a, u)
    ˞Ұൠʹ6TFSCBTFEL//$'ͱݺ͹ΕΔࣜ
    pa
    , pi
    ؒͷྨࣅ౓
    ࢀߟจݙ

    View Slide

  21. ڠௐϑΟϧλϦϯάʹΑΔ࿦จਪન <.D/FF$4$8`>
    • ྻϕΫτϧؒʢʹඃҾ༻࿦จؒʣͷྨࣅ౓Λར༻͢Δόʔδϣϯ΋͋Δɻ

    … p50
    p51
    p52
    p53
    p54 …
    p1

    p2

    p3

    p4

    p5








    ΫΤϦ
    ࿦จ
    TJNJMBS
    AAACtnichVHNShxBEP6cJP6sUdd4CeQyZDEoyNIrohIILMklp7C7ZlVwdJgZe9fW+WO6d2Ed5gXyAh48GQhBfIA8QC55AQ978JoQclTIJYfUzA6EKGo13V31ffVVV3fboSukYqw/pD14+Gh4ZHSsMP54YnKqOP1kXQadyOFNJ3CDaNO2JHeFz5tKKJdvhhG3PNvlG/bBm5Tf6PJIisB/r3oh3/asti9awrEUQWaxXjNja0Ek+ivdaEWWExuy45nxviF8/Z0pkiil9f3EUMLjUpfCmxMUzyfXEwcMEWaxxMosM/2mU8mdEnKrBcXPMLCLAA468MDhQ5HvwoKksYUKGELCthETFpEnMp4jQYG0HcrilGERekBrm6KtHPUpTmvKTO3QKS7NiJQ6Ztk5O2WX7Bs7Yz/Zn1trxVmNtJce7fZAy0Nz6sPTtd/3qjzaFfb+qe7sWaGF1axXQb2HGZLewhnou4dHl2svG7PxC/aR/aL+T1iffaUb+N0r51OdN45RoA+oXH/um876YrmyXF6qL5Wqr/OvGMUzPMccvfcKqniLGpp07hdc4Dt+aKvajsa19iBVG8o1M/jPtPAv6m2pYw==
    Pa,i =
    P
    j2Ni
    ra,j
    ⇥ sim(i, j)
    P
    j2Ni
    sim(i, j)
    ˞Ұൠʹ*UFNCBTFEL//$'ͱݺ͹ΕΔࣜ
    Lۙ๣
    Lۙ๣͔Βͷ౤ථ ֤ۙ๣ͱͷྨࣅ౓
    ࢀߟจݙ
    • ͍ͣΕͷ$'΋(PPHMFݕࡧʹൺ΂ͯ৽ͨͳ࿦จΛൃݟ͢Δͷʹ໾ཱͬͨɻ

    View Slide

  22. ڠௐϑΟϧλϦϯάͰ༻͍Δߦྻͷվྑ <-JV*&&&"DDFTT >
    • ଞͷҾ༻ݩจݙ͔Βજࡏతͳؔ܎Λਪఆ͢Δ$POUFYUCBTFE$PMMBCPSBUJWF'JMUFSJOHΛఏҊɻ

    )-JV 9,POH 9#BJ 88BOH 5.#FLFMFBOE'9JB $POUFYU#BTFE$PMMBCPSBUJWF'JMUFSJOHGPS$JUBUJPO3FDPNNFOEBUJPO
    *&&&"DDFTT WPM QQ
    p5
    p6
    p1
    p2
    p3
    p4
    p1
    ͱ p4
    ͸ಉ͡࿦จΛҾ༻
    ͍ͯ͠ͳ͍͕ɺp2
    p3
    ͔Β
    ؔ࿈ੑΛݟ͍ͩͤΔ
    p2
    ͔ΒҾ༻ p2
    ͔ΒҾ༻
    ͞Ε͍ͯͳ͍
    p1
    ͔ΒҾ༻ N11
    N12
    p1
    ͔ΒҾ༻
    ͞Ε͍ͯͳ͍
    N21
    N22
    ˠ෼ׂද͔Β p1
    ͱ p2
    ͷؔ܎ͷڧ͞ΛఆྔԽ͠ɺ
    ᮢ஋ॲཧͯ͠$'༻ͷߦྻཁૉͱ͢Δɻ
    • ,%%$VQσʔληοτͰ࣮ݧͨ݁͠Ռɺී௨ͷ$'ΑΓ΋ߴਫ਼౓ͳਪનΛୡ੒ͨ͠ɻ

    View Slide

  23. ڠௐϑΟϧλϦϯάϕʔεͷ໰୊఺
    • ߦྻ͕ඇৗʹεύʔεͰ͋Γɺ͔ͭཁૉͷ༏ྼ͕ෆ໌ྎɻ

    映画1 映画2 映画3 映画4 …
    Ϣʔβ
    5 3 2
    Ϣʔβ
    3 2 3
    Ϣʔβ
    5 2 3
    ʜ
    論⽂1 論⽂2 論⽂3 論⽂4 …
    ࿦จ
    0 1 0 0
    ࿦จ
    0 0 0 0
    ࿦จ
    0 0 0 1
    ʜ
    ঎඼΍өըͷਪન ü Ϣʔβ͕ͨ͘͞Μ͍Δ
    ü ޷͖ݏ͍ͷ৘ใ͕͋Δ
    ࿦จਪન ü Ұͭͷ࿦จ͔Βͷ౤ථ͕গͳ͍
    ü ೋ஋Ͱ͋ΓҾ༻Λશͯ౳͘͠ѻ͏

    View Slide

  24. ࿦จਪનख๏ͷΧςΰϥΠζ
    ڠௐϑΟϧλϦϯάख๏ ίϯςϯτϕʔεख๏
    άϥϑຒΊࠐΈख๏ ϋΠϒϦου

    Ϣʔβͷڵຯ ࿦จͷ಺༰
    4JNJMBS
    4JNJMBS
    Ҿ༻ Ҿ༻

    View Slide

  25. ίϯςϯτϕʔεͳ࿦จਪન
    • ࿦จͷλΠτϧɺΞϒετϥΫτɺΩʔϫʔυͳͲͷςΩετ৘ใ͔Βஶऀ΍࿦จͷ಺༰Λਪఆɻ

    • ύʔιφϥΠθʔγϣϯਪનͳͲ΋Α͘ఏҊ͞Ε͍ͯΔɻ
    ஫ҙ఺
    • ઐ໳ʹಛԽ͗͢͠Δͱɺ͢Ͱʹ஌͍ͬͯΔ࿦จ͹͔Γ͕ఏҊ͞ΕΔͱ͍͏೉఺͕͋Δɻ
    • ࿦จͷ࣭΍ਓؾ౓͸ແࢹ͞ΕΔ͜ͱ͕ଟ͍ɻ
    ࿦จਪન
    ݚڀऀϓϩϑΝΠϧ
    ྨࣅ౓ࢉग़
    ςΩετಛ௃

    View Slide

  26. 5'*%'ʹجͮ͘ख๏ <4VHJZBNB +$%-`>
    • ࿦จͷಛ௃நग़͸5'*%'ɺݚڀऀϓϩϑΝΠϦϯά͸5'Λ༻͍Δɻ
    • ޙऀ͸จॻͱ୯ޠ਺͕গͳ͍͜ͱͰ*%'ʹѱ͞Λ͢ΔͨΊɻ
    • ۀ੷͕ຊͷΈͷएखͱͦΕҎ֎ͷγχΞͰϓϩϑΝΠϦϯάํ๏Λมߋɻ
    एखɿࣗ෼ͷۀ੷ͱͦͷࢀߟจݙΛ༻͍Δɻ
    γχΞɿࣗ෼ͷۀ੷ͱͦͷࢀߟจݙʹՃ͑ͯࣗ෼ͷۀ੷ΛҾ༻͍ͯ͠ΔจݙΛ༻͍Δɻ
    • ࿦จͱݚڀऀϓϩϑΝΠϧؒͷίαΠϯྨࣅ౓Λࢉग़͠ɺ্Ґͷ࿦จΛਪનɻ

    , 4VHJZBNBBOE.: ,BO l4DIPMBSMZQBQFSSFDPNNFOEBUJPOWJBVTFSbTSFDFOUSFTFBSDIJOUFSFTUT z JO1SPD+$%- QQ
    r

    Ҿ༻ Ҿ༻
    ࠷ۙͷ࿦จ΄ͲॏΈΛେ͖͘
    Ҿ༻
    एख γχΞ

    View Slide

  27. ࿦จਪનͷݚڀ͕ٸܹʹ૿Ճ <"MJ 4DJFOUPNFUSJDT >
    • ಛʹ೥Ҏ߱͸จຒΊࠐΈ΍άϥϑຒΊࠐΈʢޙड़ʣؔ࿈ͷख๏͕੝ΜʹఏҊ͞Ε͍ͯΔɻ
    • ͜ͷ࿦จ͸ʮDJUBUJPOSFDPNNFOEBUJPOʯΛΫΤϦͱͯ͠࿦จΛऩू͕ͨ͠ɺʮQBQFS
    SFDPNNFOEBUJPOʯ΋ؚΊΔͱڪΒ͘΋ͬͱଟ͍ɻ

    ;"MJ *6MMBI ",IBO "6+BO BOE,.VIBNNBE l"OPWFSWJFXBOEFWBMVBUJPOPGDJUBUJPOSFDPNNFOEBUJPONPEFMT z
    4DJFOUPNFUSJDT WPM QQr
    ࠨਤͷ͏ͪਂ૚ֶशܥͷ࿦จ
    lDJUBUJPOSFDPNNFOEBUJPOzͷ࿦จ਺

    View Slide

  28. 4DJ#&35 <#FMUBHZ&./-1`>
    • $4ɾੜ෺ҩֶ෼໺ʹ͓͚Δ໿ສຊͷ࿦จϑϧςΩετͰ#&35Λֶशɻ
    • ΦϦδφϧ#&35ͷޠኮ਺ͱ4DJ#&35ͷޠኮ਺ͷΦʔόʔϥοϓ͸ɻ
    • Պֶ෼໺ͱҰൠ෼໺ͷςΩετͷؒͰ୯ޠͷग़ݱස౓ʹେ͖ͳҧ͍͕͋Δ͜ͱΛࣔͨ͠ɽ

    4DJ#&35
    #&35
    *#FMUBHZ FUBM l4DJ#&351SFUSBJOFE-BOHVBHF.PEFMGPS4DJFOUJGJD5FYU z1SPD&./-1
    • ͳ͓#&35͸จॻ಺ͷจ຺ͷΈΛֶशʹ༻͍ΔͨΊɺҟͳΔจॻؒͷؔ࿈ੑΛଊ͑
    ΔΑ͏ʹ͸܇࿅͞Ε͍ͯͳ͍ɻ

    View Slide

  29. จΤϯίʔμͷൺֱ <.PIBNFE)BTTBO 3FD4ZT`>
    • ࿦จਪનʹ͓͍ͯ#&35 &-.P 64& *OGFS4FOU 4DJ#&35Λൺֱɻ
    • #.ͰϥϯΩϯάީิΛߜ͔ͬͯΒΤϯίʔμͰϦϥϯΩϯά͢Δ৔߹ɼυϝΠϯಛԽܕͷ
    4DJ#&35Ͱ͸ͳ͘8JLJQFEJBؚΉҰൠจॻͰֶशͨ͠64&ͷํ͕ੑೳ͕ྑ͔ͬͨɻ
    .PIBNFE)BTTBO FUBM l#&35 &-.0 64&BOE*OGFS4FOUTFOUFODFFODPEFST5IFQBOBDFBGPSSFTFBSDIQBQFSSFDPNNFOEBUJPO z *O 1SPDPG"$.3FD4ZT
    -BUF#SFBLJOH3FTVMUTDPMPDBUFEXJUIUIFUI"$.$POGFSFODFPO3FDPNNFOEFS4ZTUFNT WPM QQr

    View Slide

  30. 41&$5&3<$PIBO "$-`>
    • 4DJ#&35Λ࿦จਪનͳͲ΁Ԡ༻͢ΔͨΊʹʮҾ༻ؔ܎ͷ༧ଌʯʹؔ͢Δ5SJQMFUϩεΛಋೖɻ
    • Ҿ༻ؔ܎͕༧ଌͰ͖ΔͳΒ͹࿦จؒͷؔ࿈ੑΛֶशͰ͖͍ͯΔͩΖ͏ͱ͍͏ߟ͑ɻ
    • ໿ສͷֶश༻USJQMFUTͱສͷݕূ༻USJQMFUTΛ༻͍ͯ5JUBO7(16Ͱֶशɻ
    • ໿೔ఔ౓Ͱֶश׬ྃɻ
    • طଘͷݴޠϞσϧʹൺ΂ͯෳ਺λεΫͰߴ͍ੑೳΛࣔͨ͠ɻ

    Ҿ༻
    ຒΊࠐΈઌͰ
    ۙ͘ͳΔΑ͏ʹ ຒΊࠐΈઌͰ
    ԕ͘ͳΔΑ͏ʹ
    "$PIBOFUBM l41&$5&3%PDVNFOUMFWFM3FQSFTFOUBUJPO-FBSOJOHVTJOH$JUBUJPOJOGPSNFE5SBOTGPSNFST z1SPD"$-

    View Slide

  31. ࿦จਪનख๏ͷΧςΰϥΠζ
    ڠௐϑΟϧλϦϯάख๏ ίϯςϯπϕʔεख๏
    άϥϑຒΊࠐΈख๏ ϋΠϒϦου

    Ϣʔβͷڵຯ ࿦จͷ಺༰
    4JNJMBS
    4JNJMBS
    Ҿ༻ Ҿ༻

    View Slide

  32. ֶज़৘ใ͔Βߏங͞ΕΔάϥϑ
    • ҟͳΔछྨͷϊʔυ΍Τοδ͕ࠞࡏ͢ΔωοτϫʔΫͷ͜ͱΛҟछ৘ใωοτϫʔΫ
    IFUFSPHFOFPVTJOGPSNBUJPOOFUXPSL)*/
    ͱΑͿɻ
    • ֶज़৘ใͷ৔߹͸ஶऀɺ࿦จɺॴଐɺձٞɺ৔ॴͳͲͷϊʔυ͕ߟ͑ΒΕΔɻ

    ֶज़৘ใͷ)*/
    ΤοδॏΈ͕શͯͰΤοδʹϥϕϧ͕
    ෇͘ͳΒ͹φϨοδάϥϑʹ૬౰
    ژ౎λϫʔ
    ҰൠతͳφϨοδάϥϑͷྫ
    ৔ॴ
    JTB
    ژ౎ࢢ
    JTMPDBUFEJO
    "MJDF
    WJTJUFE
    JTB

    View Slide

  33. 3%'7FD<3JTUPTLJ *48$`>
    • ηϚϯςΟοΫ΢Σϒ෼໺ͰΑ͘༻͍ΒΕΔɻ
    • φϨοδάϥϑ্Ͱ௖఺ˠΤοδˠ௖఺ˠΤοδʜͷΑ͏ʹάϥϑ΢ΥʔΫ͠ɺύεΛ୯ޠͷܥྻ
    ΁ม׵ɻXPSEWFDΛద༻ͯ͠ϊʔυͷຒΊࠐΈΛಘΔɻ

    1FUBS 3JTUPTLJ BOE)FJLP1BVMIFJN l3%'7FD3%'(SBQI&NCFEEJOHTGPS%BUB.JOJOH z*OUFSOBUJPOBM4FNBOUJD8FC
    $POGFSFODF
    ݚڀऀ"
    ࣥච
    ࿦จ
    ൃදઌ
    ࠃࡍձٞB

    View Slide

  34. 7013FD <,POH *&&&5&5$ >
    • ςΩετಛ௃ͱωοτϫʔΫߏ଄ʹج͖ͮ࿦จϊʔυͷຒΊࠐΈΛֶशɻ
    • ֤࿦จͷςΩετͱҾ༻άϥϑΛͦΕͧΕXPSEWFD TUSVDWFDͰϕΫτϧԽɻ
    • ͦΕͧΕͷϕΫτϧۭؒͰۙ๣ͱͳΔ࿦จͱͷΤοδΛݩͷҾ༻άϥϑʹ௥Ճ͢Δɻ
    • άϥϑ΢ΥʔΫͱ4LJQHSBNͰϊʔυຒΊࠐΈΛಘΔɻ

    9,POH ..BP 88BOH +-JVBOE#9V 7013FD7FDUPS3FQSFTFOUBUJPO-FBSOJOHPG1BQFSTXJUI5FYU*OGPSNBUJPOBOE
    4USVDUVSBM*EFOUJUZGPS3FDPNNFOEBUJPO *&&&5SBOT&NFSHJOH5PQJDTJO$PNQVUJOH WPM OP QQ

    View Slide

  35. 0"(#&35<-JV ,%%`>
    • 0QFO"DBEFNJD(SBQIͷҟछ৘ใωοτϫʔΫΛςΩετԽͯ͠#&35Λֶशɻ
    • શจΛ༻͍ͨࣄલֶशͷ࣌఺ͰλΠτϧͱΞϒετϥΫτͷจͷؒʹஶऀ໊ͷτʔΫϯΛೖΕ͓ͯ͘ɻ
    • τʔΫϯʹ͸ϊʔυͷछྨͷΠϯσοΫεΛ෇༩ɻྫɿςΩετ͸ɺஶऀ͸ɻ
    • %PDWFD΍4DJ#&35ʹൺ΂ͯ࿦จਪનੑೳͷ޲্Λୡ੒ɻ

    9-JV %:JO +;IFOH 9;IBOH 1;IBOH ):BOH :%POH BOE+5BOH l0"(#&355PXBSETB6OJGJFE#BDLCPOF
    -BOHVBHF.PEFMGPS"DBEFNJD,OPXMFEHF4FSWJDFT zJO1SPD,%%` QQr

    View Slide

  36. ࿦จਪનͷධՁํ๏
    • େ͖͘෼͚ͯ࣍ͷͭͷΞϓϩʔν͕͋Δɻ

    ࣮ࡍʹਓʹධՁͯ͠΋Β͏ɻ
    ϢʔβධՁ
    ΢ΣϒϓϥοτϑΥʔϜʹͯϢʔβͷڍಈΛධՁɻ
    ΦϯϥΠϯධՁ
    HSPVOEUSVUIΛઃఆͯ͠ਖ਼ղ཰ͳͲΛࢉग़ɻ
    ΦϑϥΠϯධՁ

    View Slide


  37. ϢʔβධՁ
    • ਪન݁ՌΛϢʔβʹఏࣔ͠ɺ߲໨͝ͱʹ఺਺Λ͚ͭͯ΋Β͏ɻ
    • ৽نੑɺ༗༻ੑɺख़࿅ऀ޲͚͔Ͳ͏͔ɺFUD

    ྫʣ<3BBNLVNBS *1. >
    "43BBNLVNBS 4'PP BOE/1BOH l6TJOHBVUIPSTQFDJGJFELFZXPSETJOCVJMEJOHBOJOJUJBMSFBEJOHMJTUPGSFTFBSDIQBQFSTJOTDJFOUJGJDQBQFS
    SFUSJFWBMBOESFDPNNFOEFSTZTUFNT z*OGPSNBUJPO1SPDFTTJOHBOE.BOBHFNFOU WPM OP QQr
    • ਓతίετ͕େ͖͍ɻಛʹ༷ʑͳ෼໺ͷධՁऀΛूΊΔͷ͸ࠔ೉ɻ
    • ධՁऀͷόοΫάϥ΢ϯυ΍εΩϧ͕Өڹ͢Δ໘΋ɻ
    ü ֶੜ໊ͱڭ৬һ໊ʹϝʔϧͰ࣮ݧڠྗΛґཔɻࢀՃ৚݅͸ʮҰຊҎ্ͷ࿦จΛࣥචͨ͜͠ͱʯɻ
    ü ܭ໰ʹର͠ஈ֊ͷϦοΧʔτई౓Ͱճ౴ɻ
    ü ֶੜͷํ͕ڭ৬һΑΓ΋ਪનʹؔ͢Δຬ଍౓͕༏Ґʹߴ͍͜ͱΛࣔͨ͠ɻ

    View Slide


  38. ΦϯϥΠϯධՁ
    • ௚઀ϨʔςΟϯάΛूΊΔ͜ͱͳ͘ਪનγεςϜར༻ऀͷڍಈΛఆྔԽ͢Δɻ
    • &ίϚʔεͰͷਪનΞϧΰϦζϜධՁͰΑ͘༻͍ΒΕΔํ๏ɻ
    • ΫϦοΫεϧʔ཰ʜਪન͞Εͨ΋ͷͷ͏ͪɺ࣮ࡍʹΫϦοΫ͞Εׂͨ߹ɻϢʔβͷຬ଍౓ͱ૬͕ؔ͋Δɻ
    • μ΢ϯϩʔυ཰ɺߪೖ཰ ͳͲ

    • ϓϥοτϑΥʔϜͷҧ͍͕༩͑ΔӨڹΛݕূ͢Δඞཁ͕͋Δɻ
    • ࣮γεςϜ͕ඞཁͳͨΊֶज़৘ใਪનͷ෼໺Ͱ͸ಛఆάϧʔϓ͔͠ߦ͍ͬͯͳ͍ɻ
    ྫʣ<$PMMJOT+$%-`>
    จݙਪનɾ؅ཧͷΦʔϓϯιʔεϓϩδΣΫτʹͯສຊͷจݙ͔ΒϢʔβʹਪનɻ೥൒࣮ࢪɻ
    TF-IDF
    (95%)
    Doc2vec (2%)
    キーフレーズ(3%)
    ਪનͷΞϧΰϦζϜ಺༁
    ϓϥοτϑΥʔϜʹΑͬͯ
    ΞϧΰϦζϜͷੑೳॱҐ͕มಈͨ͠ʢʂʣ

    View Slide


  39. ΦϑϥΠϯධՁ
    • ࿦จσʔληοτ಺ͷҾ༻ؔ܎ΛٖࣅతʹHSPVOEUSVUIͱͯ͠औΓѻ͏ɻ
    • ਪનͨ͠࿦จ͕ΫΤϦ࿦จʹҾ༻͞Ε͍ͯΕ͹ਖ਼ྫɺ͞Ε͍ͯͳ͚Ε͹ෛྫɻ
    • ϢʔβධՁΑΓ΋ਓతίετ͕௿͍ͨΊ΄ͱΜͲͷݚڀ͕͜ͷධՁํ๏Λ࠾༻ɻ

    ྫʣ%#-1σʔληοτ
    ຊͷ࿦จ
    ݸͷҾ༻ؔ܎
    ྫʣ"$-"OUIPMPHZ/FUXPSL
    • ʮຊ౰ͳΒҾ༻ͯ͠΋͓͔͘͠ͳ͍࿦จʯ͕ෆਖ਼ղʹͳͬͯ͠·͏ɻ
    IUUQTXXXBNJOFSPSHDJUBUJPO
    ຊͷ࿦จ
    ݸͷҾ༻ؔ܎
    IUUQTBBOIPXEPXOMPBE
    ʜͲͷධՁํ๏΋Ұ௕Ұ୹ɺ͞ΒͳΔݕ౼͕ඞཁɻ

    View Slide

  40. ࠷ۙͷݚڀɿΞεϖΫτผͷจຒΊࠐΈ
    • 4PMVUJPO5BJMPS <5BLBIBTIJ,BUTVSBJ &$*3`>
    • λεΫ͕ࣅ͍ͯΔ͕ख๏͕ҟͳΔ࿦จΛਪનɻ

    • 4QFDJBMJ[FE%PDVNFOU&NCFEEJOHT <0TUFOEPSG +$%-`>
    • ࿦จͷλεΫɺख๏ɺσʔληοτʹؔ͢Δ෦෼͝ͱʹຒΊࠐΈΛࢉग़͢Δɻ
    • 1BQFST8JUI$PEFͷ UBTL NFUIPE EBUBTFUʹؔ͢Δهड़Λ΋ͱʹUSJQMFUTΛ࡞੒ɻ
    ࿦จ ࿦จ
    ಉ͡σʔληοτΛ
    ࢖͍ͬͯΔ
    ˠۙ͘ʹ഑ஔ
    ෼໺"ͷ࿦จ ෼໺#ͷ࿦จ
    ᶃΞϒετϥΫτͷྨࣅ౓
    ᶄ໰୊จɾղܾจͷྨࣅ౓
    ྨࣅ౓ͷߴ͍ॱʹਪન
    ໰୊จ
    ղܾจ
    #&35Λ༻͍ͯΞϒετ಺ͷจΛ
    എܠɾ໨తɾख๏ʹ෼ྨ
    . 0TUFOEPSGG FUBM l4QFDJBMJ[FEEPDVNFOUFNCFEEJOHTGPSBTQFDUCBTFETJNJMBSJUZPG
    SFTFBSDIQBQFST z JO1SPD+$%-"SUJDMF QQr
    55BLBIBTIJBOE.,BUTVSBJ l4PMVUJPO5BJMPS4DJFOUJGJD1BQFS3FDPNNFOEBUJPO
    #BTFEPO'JOF(SBJOFE"CTUSBDU"OBMZTJT zJO&$*3 QQr

    View Slide

  41. ·ͱΊͱࠓޙͷల๬
    • ֶज़৘ใݕࡧͱਪનʹؔͯ͠ɺಛʹ࿦จΛର৅ͱͨ͠ݚڀΛ঺հͨ͠ɻ
    • ਪન͸ΦϑϥΠϯධՁ͕ओྲྀ͕ͩɺਖ਼ղσʔλͷ࡞Γํʹ͸͞ΒͳΔݕ౼͕ඞཁɻ
    • શจ͕࢖͑Δ࿦จσʔλϕʔε͸ݶΒΕ͓ͯΓɺ͔ͭ෼໺ԣஅతͳֶज़৘ใਪનʹ
    ؔ͢Δݚڀ͸ൃల్্ͱ͍͑Δɻ
    • ࣗવݴޠॲཧʹΑΔΞεϖΫτ෼ੳ͸ࠓޙ͞Βʹల։͍ͯ͘͠ͱ༧૝ɻ

    ڞಉݚڀͳͲ͓ؾܰʹ͓੠͕͚͍ͩ͘͞ʂ
    ؔ࿈ݚڀऀਪન ݚڀτϨϯυ෼ੳ ஶऀಉఆ
    ৘ใܥ
    ෺ཧܥ
    ҩֶܥ
    ը

    ҩ

    ϩ
    Ϙ
    ο
    τ
    ͦͷଞͷݚڀτϐοΫ

    View Slide