The first step of the Top-k Selection algorithm in search

Transcript

1. The first step of
   the Top-k Selection
   algorithm in search
   M3 tecktalk
   #164
   hiromu-nakamura @po3rin
   

   View Slide

2. தଜ߂෢
   ΤϜεϦʔͷ"*ɾػցֶशνʔϜͰݕࡧج൫पΓ
   Λ୲౰͍ͯ͠Δɻ(P͕޷͖ɻ
   ֤छΞΧ΢ϯτQPSJOͰ΍ͬͯ·͢ɻ
   )*SPNV/BLBNVSB
   

   View Slide

3. View Slide

4. ݕࡧΤϯδϯ͸ΫΤϦΛड͚औΓɺυΩϡϝϯτ͝ͱʹ
   είΞܭࢉΛείΞ্ҐL݅Λฦ͢ɻ
   4FBSDI&OHJOF
   ΫΤϦ
   L݅ͷείΞॱυ
   Ωϡϝϯτ
   

   View Slide

5. &MBTUJDTFBSDIͳͲͷݕࡧΤϯδϯ͸ɺ୯ޠͷҐஔ৘ใΛ
   ֨ೲ͢ΔͨΊͷࡧҾσʔλߏ଄Ͱ͋ΔసஔΠϯσοΫε
   Λ࠾༻͍ͯ͠Δɻ
   4FBSDI&OHJOF
   %ࢲ͸ম೑Λ৯΂Δ
   %ࢲ͸αϥμ΋৯΂Δ
   ʮম೑ʯ
   %
   సஔΠϯσοΫε
   

   View Slide

6. ݕࡧʹ͓͚ΔυΩϡϝϯτͷείΞϦϯάʹ͸0LBQJ
   #.ͳͲ୅දతͳ΋ͷ͕͋ΔLɻ͜ΕΛݕࡧʹώοτ
   ͢ΔυΩϡϝϯτʹରͯ͠ܭࢉ͍ͯ͘͠ɻ
   4DPSJOH
   

   View Slide

7. υΩϡϝϯτશͯͷείΞϦϯά͕ऴ͔ྃͯ͠ΒͦΕΒΛฒ
   ͼସ͑Δͷ͸ϥϯΩϯά্ҐL൪໨·Ͱͷ݁Ռʹڵຯ͕͋Δ৔
   ߹ඇޮ཰ɻͦ͜Ͱݕࡧʹ͓͚ΔLݸͷ࠷େཁૉʢ·ͨ͸࠷খ
   ཁૉʣΛબ୒͢Δ໰୊Λߟ͑Δɻ
   5PQLTFMFDUJPO
   

   View Slide

8. 5PQL
   TFMFDUJPO
   ೋ෼ώʔϓʹυΩϡϝϯτͱείΞΛอ͍࣋ͯ͘͜͠ͱͰ࠷
   ऴతͳιʔτͷίετΛεϧʔͰ͖Δɻώʔϓʹ֨ೲ͢Δ͔
   ͸ࠜͷείΞ͚ͩݟΕ͹ྑ͍ɻ
   ೋ෼ώʔϓ
   
   
   
   

   View Slide

9. 5PQL
   TFMFDUJPO
   ࠓ·Ͱݕࡧʹώοτͨ͠શͯͷυΩϡϝϯτͷείΞΛܭࢉ
   ͍͕ͯͨ͠ɺ࣮͸είΞΛܭࢉ͠ͳͯ͘΋ɺ͋ΔυΩϡϝϯ
   τ͕UPQLʹೖΒͳ͍͜ͱΛ֬ఆͰ͖Δํ๏͕͋ΔɻͦΕ͕
   NBYTDPSFͱݺ͹ΕΔख๏Ͱ͢ɻ
   .BYTDPSF
   

   View Slide

10. ͷ໾ׂ͸ΫΤϦλʔϜͦΕͧΕͷείΞ΁ͷߩݙ౓ͷ্ݶ
    Λܾఆ͢Δ͜ͱɻ ͷܦݧతॳظ஋Λ࠾༻͢Δͱɺ5'஋
    ্ࣜͷୈҼࢠ
    ͸ ͭ·ΓΛ௒͑ͳ͍ɻ*%'΋ࣄલ
    ʹܭࢉͰ͖Δɻ͜ΕʹΑΓ֤λʔϜ͕د༩͢Δ࠷େείΞ͸
    ࣄલʹܾఆͰ͖ΔɻͦͷͨΊϥϯΩϯά্ҐʹݱΕͳ͍υ
    Ωϡϝϯτ͸͋Β͔͡ΊΘ͔Δɻ
    k1
    k1
    k1
    + 1
    4DPSJOH
    5PQL
    TFMFDUJPO
    

    View Slide

11. 4DPSJOH
    λʔϜ .BY4DPSF
    UIF
    TFBSDI
    FOHJOF
    
    
    
    ͜ͷ࣌఺ͰzUIFzؚ͔͠·ͳ͍ϙεςΟϯάϦετͷείΞܭ
    ࢉ͸ແࢹͰ͖ΔɻͳͥͳΒ࠷େείΞ͸ݱࡏͷώʔϓͷࠜ
    ͷείΞΛ௒͑ͳ͍͔Βɻ
    5PQL
    TFMFDUJPO
    

    View Slide

12. 4DPSJOH
    λʔϜ .BY4DPSF
    UIF
    TFBSDI
    FOHJOF
    
    
    
    ͜ͷ࣌఺Ͱ͸zUIFzͱzTFBSDIzؚ͔͠·ͳ͍υΩϡϝϯτͷε
    ίΞܭࢉ͸ແࢹͰ͖ΔɻͳͥͳΒ࠷େείΞ͸ݱࡏͷ
    ώʔϓͷࠜͷείΞΛ௒͑ͳ͍͔Βɻ
    5PQL
    TFMFDUJPO
    

    View Slide

13. *OUFSWBMCBTFEQSVOJOH
    TFBSDI
    FOHJOF
    %PDVNFOU
    ◾◾◾◾◾◾
    ◾◾◾◾◾
    NBY
    NBY
    
    
    TDPSF
    ͜ͷ࣌఺ͰzFOHJOFz͚͔ͩ͠ग़ݱ͍ͯ͠ͳ͍υΩϡϝϯτ͸ແࢹͰ͖Δ
    ͜ͷ࣌఺ͰzTFBSDIz͚͔ͩ͠ग़ݱ͍ͯ͠ͳ͍υΩϡϝϯτ͸ແࢹͰ͖Δ
    XBMLEPDT
    

    View Slide

14. *OUFSWBMCBTFEQSVOJOH
    ࣮ફͰ͸ϙεςΟϯάϦετ͸ѹॖ͞Ε͍ͯΔͷͰϒϩοΫ
    αΠζ͝ͱʹNBYTDPSFΛอ࣋͠ ֤ϒϩοΫͷ۠੾ΓͷΠϯ
    λʔόϧʹ6QQFS#PVOE4DPSF 6#4
    Λܭࢉ͓ͯ͘͠ɻ͜Ε
    ʹΑΓϒϩοΫղౚΛεΩοϓͰ͖Δɻ
    TFBSDI
    FOHJOF
    %PDVNFOU
    ◾◾◾◾◾◾
    ◾◾◾◾
    NBY
    NBY NBY NBY
    NBY
    
    6#4
    

    View Slide

15. $PODMVTJPO
    5PQLTFMFDUJPOͰ͸ೋ෼ώʔϓͳͲͷσʔλߏ଄͕ར༻Ͱ͖Δɻ
    ݕࡧείΞʹ͓͚Δ֤λʔϜ͕ߩݙ͢ΔNBYTDPSF͸ࣄલʹܭࢉͰ͖ΔͷͰείΞܭࢉΛεΩοϓͰ͖Δɻ
    *OUFSWBMCBTFEQSVOJOHΛར༻͢Δ͜ͱͰѹॖΠϯσοΫε΋είΞܭࢉΛޮ཰తʹߦ͑Δɻ
    -VDFOFͰ͸ߋʹൃలܥͷ8"/%ͱ͍͏ΞϧΰϦζϜ͕ಋೖ͞Ε͍ͯΔɻ
    

    View Slide

16. ࢀߟ
    -VDFOFͷ5PQLΫΤϦϓϩηογϯά࠷దԽ
    .BY4DPSFΞϧΰϦζϜฤ
    IUUQTMJOLNFEJVNDPN&I%"D$HRKDC
    0LBQJ#.
    IUUQTKBXJLJQFEJBPSHXJLJ0LBQJ@#.
    ৘ใݕࡧݕࡧΤϯδϯͷ࣮૷ͱධՁ
    UUQTXXXBNB[PODPKQEQSFGDN@TX@FN@S@NU@EQ@$H'C&;7('9
    

    View Slide

17. The first step of
    the Top-k algorithm
    in search
    M3 tecktalk
    #164
    hiromu-nakamura @po3rin
    

    View Slide