トピックモデル第2章

Transcript

1. (VOPTZσʔλϚΠχϯάݚڀձ
   
   τϐοΫϞσϧʹΑΔ౷ܭతજࡏҙຯղੳ
   -BUFOU%JSJDIMFU"MMPDBUJPO
   ৿ຊ३࢘
   

   View Slide

2. -BUFOU%JSJDIMFU"MMPDBUJPO
   ֓ཁ
   -%"ͷੜ੒աఔ
   ଟ߲෼෍ͱ%JSJDIMFU෼෍
   -%"ͷزԿֶతղऍ
   -%"ͷԠ༻ྫ
   આ໌ͷ౎߹্͋͑ͯٯస
   

   View Slide

3. ֓ཁ
   w -BUFOU%JSJDIMFU"MMPDBUJPO -%"
   ౷ܭతજࡏҙຯղ
   ੳΛՄೳʹ͢Δ౷ܭϞσϧ
   w #BHPG8PSETͰ୯ޠͱग़ݱස౓ͷू߹ΛϞσϧԽ
   w #P8Ͱ͸౷ܭతʹڞى͠΍͍͢୯ޠΛදݱͰ͖Δʢ͸ͣʣ
   w -%"͸ڞىੑΛ਺ཧతʹදݱ͢Δ
   w จॻʹ͸જࡏతͳʢ௚઀؍ଌͰ͖ͳ͍ʣτϐοΫ͕͋Δ
   w જࡏτϐοΫΛ؍ଌͰ͖ͳ͍֬཰ม਺ͰఆࣜԽ
   w w w w w w w w w w
   w Ұͭͷจॻʹෳ਺ͷજࡏτϐοΫ͕ଘࡏ͢Δ
   w Ԡ༻ൣғ͸޿͍
   w Ϣʔβʔͷߪങཤྺ#BHPG*UFNT
   w ϢʔβʔͷهࣄӾཡཤྺ#BHPG/FXTBSUJDMFT
   

   View Slide

4. w ୯ޠͷछྨ਺7
   w ୯ޠͷΠϯσοΫεू߹
   w τϐοΫLʹ͓͚Δ୯ޠWͷग़ݱ֬཰
   w ୯ޠͷग़ݱ෼෍
   w จॻʹ͸༷ʑͳτϐοΫʢܦࡁɺεϙʔπɺܳೳʣ͕ଘࡏ
   w τϐοΫ͝ͱʹ୯ޠͷग़ݱස౓͕ҟͳΔ㲗୯ޠͷग़ݱස౓͕τϐοΫͷಛ௃Λද͢
   w ྫʣ੓࣏ʹग़ͯ͘Δʮ೔ຊʯεϙʔπʹग़ͯ͘Δʮ೔ຊʯ
   w ͲͷΑ͏ʹ਺ཧతʹϞσϧԽ͢Δ͔ʢม਺Λ੔ཧʣ
   જࡏτϐοΫͷ਺ཧϞσϧԽ
   -%"ੜ੒աఔ
   τϐοΫ͸؍ଌͰ͖ͳ͍જࡏτϐοΫͱͯ͠ϞσϧԽ
   ˠજࡏτϐοΫ͝ͱͷ୯ޠͷग़ݱ෼෍ʹΑͬͯදݱ
   

   View Slide

5. w lQMBZz͸ෳ਺ͷҙຯΛ࣋ͭ
   w εϙʔπɺԋܶɺԻָ
   w ਓ͕͜ΕΒͷҧ͍Λཧղ͢Δͷ͸༰қ
   w -%"Ͱ΋ಉ༷ʹҧ͍ΛදݱͰ͖Δˠʮ౷ܭతજࡏҙຯղੳʯ
   w τϐοΫͷજࡏతҙຯΛද͢ͷ͸ɺ୯ޠग़ݱ֬཰П
   w ग़ݱ֬཰ͷߴ͍ॱʹ୯ޠΛݟΕ͹τϐοΫ͕Θ͔Δ
   જࡏม਺ʹΑΔଟٛੑͷϞσϧԽ
   -%"ੜ੒աఔ
   

   View Slide

6. จॻ਺. จॻEͷ૯୯ޠ਺OE จॻ͸ෳ਺ͷτϐοΫ͔Βߏ੒ τ
   ϐοΫ෼෍В ୯ޠͷग़ݱ෼෍П
   В П͸֬཰ϕΫτϧˠੜ੒͸֬཰ϕΫτϧ্ͷ֬཰෼෍Ͱ͋Δ
   %JSJDIMFU෼෍ʹै͏ͱԾఆ
   ఆࣜԽͱάϥϑΟΧϧϞσϧ
   -%"ੜ੒աఔ
   ୯ޠX જࡏτϐοΫ[͸཭ࢄ஋ͷͨΊɺଟ߲෼෍Λੜ੒෼෍ͱԾఆ
   Qਤ-%"ͷάϥϑΟΧϧϞσϧΛࢀর
   

   View Slide

7. w ଟ߲෼෍ͷڞ໾ࣄલ෼෍͋Γɺ਺ֶతʹऔΓѻ
   ͍΍͍͢
   w จॻதͷ୯ޠͷग़ݱස౓͸ଟ߲෼෍ʹै͏
   w ڞ໾ɿ̎ͭΛೖΕ͔͑ͯ΋ੑ࣭ʹมԽ͕ى͜Βͳ͍Α͏ͳؔ܎
   w ୯ޠग़ݱ֬཰ͷภΓѻ͍΍͍͢Ϟσϧ
   w ͋ΔจষͰ୯ޠ͕ͰΔ֬཰͸Ұ༷Ͱ͸ͳ͘ɺภ͍ͬͯΔ
   ͳͥ%JSJDIMFU෼෍Λ࠾༻͔ͨ͠
   ଟ߲෼෍ͱ%JSJDIMFU෼෍
   

   View Slide

8. w දͷग़Δ֬཰͕КͷίΠϯΛ/ճ౤͛ͨ࣌ɺද͕ग़Δճ਺Oͷ֬཰෼෍
   ೋ߲෼෍ͱଟ߲෼෍
   ଟ߲෼෍ͱ%JSJDIMFU෼෍
   w OճͷࢼߦͰ֤஋ͷग़ݱճ਺OLʹڵຯ͕͋Δ৔߹ͷ֬཰෼෍
   

   View Slide

9. w ֊૚ϕΠζϞσϧͱݺ͹ΕΔ֊૚తͳ֬཰Ϟσϧ
   w Кʹؔ͢Δ֬཰෼෍
   %JSJDIMFU෼෍
   ଟ߲෼෍ͱ%JSJDIMFU෼෍
   w ΨϯϚؔ਺ͷੑ࣭
   

   View Slide

10. w Q Y
    ΛQ ZcY
    ͷࣄલ෼෍ͱ͢Δɻࣄޙ෼෍Q YcZ
    㲍Q ZcY
    Q Y
    ͷ෼෍͕
    Q Y
    ͱಉ͡෼෍ʹͳΔͱ͖ɺQ Y
    ͸Q ZcY
    ͷڞ໾ࣄલ෼෍Ͱ͋Δͱ͍͏
    ॏཁʂ͓֮͑ͯ͜͏
    ଟ߲෼෍ͱ%JSJDIMFU෼෍
    w ม਺ͷҙຯ߹͍͸ҟͳΔଟ߲෼෍ͱ%JSJDIMFU෼෍ͷߏ଄͸Α͘
    ࣅ͍ͯͯɺ਺ֶతͳऔΓѻ͍΍༻ҙʹͳΔ
    w ࣄલ෼෍ʹ͓͚ΔЋL͸ࣄޙ෼෍Ͱ͸LͷݱΕͨස౓OLʹΑͬͯ
    ՃࢉɻЋL͸σʔλΛ؍ଌ͢ΔલͷLͷԾ૝తස౓Λࣔ͢
    

    View Slide

11. w ୯ޠ QMBZ HBNF NVTJD
    ͔͠ͳ͍จॻ
    w จॻ"ʹ͓͚Δ୯ޠස౓G" QMBZ HBNF NVTJD
    
    
    w จॻ#ʹ͓͚Δ୯ޠස౓G# QMBZ HBNF NVTJD
    
    
    w ୯ޠ෼෍Q"
    Q#
    
    ˠ୯ޠ࠲ඪ୯ମ্΁จॻΛࣹӨ
    -%"ͷ࣍ݩѹॖɾ௿࣍ݩۭؒ΁ͷࣹӨ
    -%"ͷزԿֶతղऍ
    NVTJD
    HBNF QMBZ
    จॻ͸఺ ୯ޠ࠲ඪ୯ମ
    

    View Slide

12. w જࡏτϐοΫ࠲ඪ୯ମΛܗ੒͢ΔجఈϕΫτϧ͸୯ޠ࠲ඪ୯ମ্ͷ఺
    w جఈϕΫτϧ͸୯ޠϕΫτϧ
    w П͸જࡏτϐοΫ࠲ඪ୯ମΛܗ੒͢ΔجఈϕΫτϧ
    w ୯ޠ࠲ඪ୯ମͷ఺
    w จॻ͝ͱͷτϐοΫ෼෍В͸જࡏτϐοΫ࠲ඪ୯ମ্ͷ఺
    w -%"͸จॻू߹ͷજࡏτϐοΫ୯ମ্΁ͷࣹӨ
    w જࡏτϐοΫ࠲ඪ୯ମ͸୯ޠ࠲ඪ୯ମΑΓ΋௿࣍ݩˠ࣍ݩѹॖ
    w -%"Ͱ͸%JSJDIMFU෼෍ʹ͕ͨͬͯ͠จॻΛજࡏτϐοΫ࠲ඪ୯ମ্ʹ
    ഑ஔ
    ˠ-BUFOU%JSJDIMFU"MMPDBUJPO
    જࡏτϐοΫ࠲ඪ୯ମ͸୯ޠ࠲ඪΑΓ΋௿࣍ݩ
    -%"ͷزԿֶతղऍ
    

    View Slide

13. w τϐοΫΛ஥հͯ͠ɺจॻதʹग़ݱ͠ͳ͍୯ޠͷ
    ֬཰Λࢉग़Ͱ͖Δ
    w จॻதʹݱΕΔ୯ޠͷස౓ͷΈͰܭࢉ͢Δͱɺग़ݱ͠ͳ͍୯ޠ
    ʹ֬཰͸θϩʹͳͬͯ͠·͏
    w τϐοΫΛ஥հ͠ɺपғͷ୯ޠ৘ใΛݩʹ୯ޠWͷ֬཰Λࢉग़Ͱ
    ͖Δ
    w هड़தͷจॻͰ࢖༻Ͱ͖Δ୯ޠ΍ɺݴ͍׵͑ͷఏࣔ
    ༧ଌ
    -%"ͷԠ༻
    

    View Slide

14. w ৘ใݕࡧΫΤϦͰΑΓؔ࿈౓߹ͷߴ͍΋ͷΛݕࡧ
    w ΫΤϦRʹؚΜͰ͍ͳ͍จॻʹରͯ͠΋֬཰ΛܭࢉͰ͖Δ
    w DBSΛؚΜͩΫΤϦʹର͠ɺBVUPNPCJMFؚ͕·ΕΔจॻ΋༗ޮʹʢڞىੑΛ༗ޮࢹʣ
    w Q DBScL
    ͱQ BVUPNPCJMFcE
    ͕ಉఔ౓ͷ֬཰ͷ৔߹
    w ୯ޠͷग़ݱ͠΍͕͢͞৘ใݕࡧͷੑೳͷྑ͞Ͱ͸ͳ͍
    w ਪનγεςϜ΁ͷԠ༻
    w #BHPG*UFNTͰڞى͢ΔΞΠςϜΛਪન
    w τϐοΫϢʔβͷજࡏతᅂ޷ύλʔϯɹ
    ༧ଌ
    -%"ͷԠ༻
    

    View Slide

15. w ϥϕϧࣝผث
    w ֤୯ޠ͔Βજࡏม਺ͷ஋ͱೖΕସ͑Δͱɺ࣍ݩѹॖͨ͠จॻΛදݱͰ͖Δ
    w ࣍ݩѹॖͨ͠จॻ͔Βಛ௃ϕΫτϧநग़
    w ڭࢣϥϕϧ෇͖σʔλ͔Β෼ྨثΛ࡞੒
    w ද໘తʹݱΕͳ͍જࡏతʹؔ࿈͢Δ୯ޠͰ෼ྨͰ͖Δ
    w ػց຋༁
    w ຋༁ʹ༻͍Δ୯ޠީิʹจ຺৘ใΛར༻͍ͯ͠ͳ͍ͨΊɺ͠͹͠͹ෆࣗવͳ୯ޠͳ
    ୯ޠ͕બ͹ΕΔ
    w τϐοΫ৘ใΛར༻͢Δ͜ͱͰɺ୯ޠͷީิΛߜΓΑΓࣗવͳ຋༁จΛ࡞੒Ͱ͖Δ
    જࡏ৘ใͷར༻
    -%"ͷԠ༻
    

    View Slide

16. ͓͠·͍
    

    View Slide