クックパッドにおける研究の活用と支援

558ffd4e5cf6c95e60c91333677b93c4?s=47 j.harashima
June 20, 2015
7.1k

 クックパッドにおける研究の活用と支援

558ffd4e5cf6c95e60c91333677b93c4?s=128

j.harashima

June 20, 2015
Tweet

Transcript

  1. ΫοΫύουʹ͓͚Δݚڀͷ׆༻ͱࢧԉ ΫοΫύουגࣜձࣾ ݪౡ७

  2. ൃද֓ཁ ΫοΫύουͰ͸ɺ࠷৽ͷݚڀΛΩϟονΞοϓͯ͠ϓϩμΫτʹ׆༻͢Δ ͱͱ΋ʹɺֶज़తͳݚڀΛαϙʔτͯ͠ΞΧσϛΞʹߩݙ͢Δ͜ͱʹ΋ྗΛ ೖΕ͍ͯ·͢ɻຊൃදͰ͸ɺΫοΫύουʹ͓͚Δ࠷৽ͷݚڀͷ׆༻ࣄྫͱ ֶज़తͳݚڀΛࢧԉ͢ΔͨΊͷऔΓ૊Έʹ͍ͭͯ঺հ͠·͢ɻ ΫοΫύου ݚڀ

  3. ݪౡ७ ɾΤϯδχΞʢݕࡧɾฤ੒෦ʣ ɹɾ ZFBSTPGEFWFMPQNFOUFYQFSJFODF ɾത࢜ʢ৘ใֶʣ ɹɾ ZFBSTPGSFTFBSDIFYQFSJFODF ɾઐ໳ ɹɾࣗવݴޠॲཧɺ৘ใݕࡧ

  4. ΫοΫύου

  5. ར༻ঢ়گ IUUQTJOGPDPPLQBEDPNpOBODJBM@IJHIMJHIU

  6. Ϩγϐ਺ IUUQTJOGPDPPLQBEDPNpOBODJBM@IJHIMJHIU

  7. ΫοΫύουʹ͓͚Δݚڀͷ׆༻ࣄྫ

  8. ΫοΫύουʁ ʮ3VCZPO3BJMTͱ͔ɺΤϯδχΞϦϯά͕Μ͹ͬͯͦ͏ɻ ɹ͚Ͳɺݚڀʹ͸ڵຯͳ͍Ͱ͠ΐʁʯ

  9. ΫοΫύουͷݚڀࣄ৘ ɾࠓ݄ɺݚڀྖҬΛઐ໳ʹѻ͏νʔϜ͕ൃ଍͠·ͨ͠ ɹɾ੒Ռ͸͜Ε͔Β ɾ༷ʑͳݚڀΛϓϩμΫτʹ׆༻͍ͯ͠·͢ ɹɾϨγϐͷࣗಈ෼ྨ ɹɾ঎඼ͷݻ༗දݱೝࣝ ɹɾίϯςϯπͷϨίϝϯσʔγϣϯ

  10. Ϩγϐͷࣗಈ෼ྨ

  11. Ϩγϐͷࣗಈ෼ྨ

  12. ɾ47.ʹΑΔ෼ྨ ɹɾ-*#-*/&"3Λར༻ ɹɾૉੑ͸ൿີ ɾϢʔβʹΑΔਪન݁ՌΛ׆༻ ɹɾਖ਼ղσʔλΛߏங͢ΔखؒΛলུ ɾຖ೔શϨγϐʢ໿ສ඼ʣΛ෼ྨ FHεʔϓͱͯ͠ਪન͞ΕͨϨγϐΛɹ ɹ ɹ εʔϓͷਖ਼ղσʔλͱͯ͠ར༻

    Ϩγϐͷࣗಈ෼ྨ
  13. Ϩγϐͷࣗಈ෼ྨ

  14. ϖʔδԼʹ ϑΟϧλʔ Ϩγϐͷࣗಈ෼ྨ

  15. ঎඼ͷݻ༗දݱೝࣝ

  16. ಛച৘ใʢళฮ͔Β౤ߘʣ ݻ༗දݱೝࣝ ঎඼ͷݻ༗දݱೝࣝ ͨ·Ͷ͗ݸ ԁ ͨ·Ͷ͗υϨογϯά ԁ   ͨ·Ͷ͗ݸ

    ਺ྔ ঎඼໊ ͨ·Ͷ͗υϨογϯά ঎඼໊ ʮͨ·Ͷ͗ʯͰΠϯσΩγϯά ʮͨ·Ͷ͗υϨογϯάʯͰΠϯσΩγϯά 0OMJOF Ϣʔβͷۙลͷళฮ͔Β ঎඼໊ʮͨ·Ͷ͗ʯΛݕࡧ
  17. ɾ$3'ʹΑΔݻ༗දݱೝࣝ ɹɾ$3'TVJUFΛར༻ ɾ໿छྨͷϥϕϧΛఆٛ ɹɾFH঎඼໊ɺ࢈஍ɺ਺ྔɺ෦Ґ ɾਓखͰਖ਼ղσʔλΛߏங ɹɾܗଶૉղੳͱݻ༗දݱೝࣝͷਖ਼ղσʔλΛߏங ɹɾ֤ϞσϧΛ঎඼ʹ࠷దԽ ঎඼ͷݻ༗දݱೝࣝ

  18. ݻ༗දݱೝࣝͷਖ਼ղ ܗଶૉղੳͷਖ਼ղ ঎඼ͷݻ༗දݱೝࣝ

  19. ઍ༿࢈ʹΜ͡Μຊ ࠃ࢈ಲ೑ʢϞϞɾΧλʣ ࢈஍ ਺ྔ ઍ༿࢈ʹΜ͡Μຊ ࠃ࢈ಲ೑ʢϞϞɾΧλʣ ෦Ґ ঎඼໊ ෦Ґ ࢈஍

    ঎඼໊ ঎඼ͷݻ༗දݱೝࣝ ͚ͨͷ͜ͷཬ ͚ͨͷ͜ͷཬ ঎඼໊
  20. ΫΤϦʹԠͯ͡࠷దͳίϯςϯπΛදࣔ ίϯςϯπͷ࠷దԽ

  21. ίϯςϯπͷ࠷దԽ ݙཱΛग़ͨ͠Βศརʁ Ϣʔβ͞ΜʹΑΔ ͓͢͢ΊϨγϐͱ͔ ྉཧͷίπͳΜͯͲ͏ʁ 6*͕ෳࡶԽ ৽͍͠ίϯςϯπΛग़ͦ͏ʂ χϡʔεɺಈըɺ

  22. ɾଟ࿹όϯσΟοτ ɹɾ୳ࡧͱ׆༻Λ࢖͍෼͚ͯใुΛ࠷େԽ ɹɹɾ୳ࡧɿ৘ใΛऩू͢ΔͨΊʹ࿹Λબ୒ ɹɹɾ׆༻ɿऩूͨ͠࿹ʹج͍ͮͯ࿹Λબ୒ ɹɾ୳ࡧͱ׆༻͸τϨʔυΦϑ ɾίϯςΩετ෇͖ଟ࿹όϯσΟοτ ɹɾ࿹ͷใु͕ίϯςΩετʹΑͬͯҧ͏ʢFH޿ࠂͷ$53ʣ ɹɾίϯςΩετΛૉੑͰදݱͯ͠ɺใु͕࠷େͱͳΔΑ͏ʹॏΈΛֶश ɹɾίϯςΩετ͝ͱʹ࿹Λબ୒ ࿹

    ίϯςϯπͷ࠷దԽ
  23. ɾίϯςΩετ෇͖ଟ࿹όϯσΟοτʹΑΔίϯςϯπ࠷దԽ ɹɾ7PXQBM8BCCJUΛར༻ ɾ࿹ʹίϯςϯπ ɹɾFHݙཱɺهࣄɺྉཧಈը ɾใुʹ$53 ɹɾϢʔβʹظ଴͞ΕΔ࿹͕׆༻࣌ʹબ୒͞ΕΔ ɾ࿹ͷબ୒ ɹɾ୳ࡧ࣌ɿҰ෦17ͰϥϯμϜʹબ୒ ɹɾ׆༻࣌ɿ$53࠷େͱظ଴͞ΕΔ΋ͷΛબ୒ 10/9

    10/10% 10/11% 10/9 10/10% 10/11% m/d m/d md%   % md%  % construct%training% and%test%data m/d% % %m’/d’%  mdm’d’ %   % Amazon%S3    10/9% % 10/11%  … … predict%ac:ons learn%model%     %  %  Offline  explore exploit log*show/click*events  Treasure*Data           Online ίϯςϯπͷ࠷దԽ
  24. ɾ$53ഒ ɹɾ ׆༻࣌ͷ$53  ୳ࡧ࣌ͷ$53 ɾબ୒͞Εͨίϯςϯπʢׅހ಺͸֤ΫϦΤΠςΟϒͷ$53ʣ ྫྷౚཛ هࣄʢʣ ͚ͨͷ͜ ๐ཱͯੜΫϦʔϜ؆୯

    جຊ஌ࣝʢʣ ྉཧಈըʢʣ ίϯςϯπͷ࠷దԽ
  25. ɾΫοΫύουʹ͓͚Δݚڀͷ׆༻ࣄྫ ɹɾϨγϐͷࣗಈ෼ྨ ɹɾ঎඼ͷݻ༗දݱೝࣝ ɹɾίϯςϯπͷ࠷దԽ ɾࠓޙͷల๬ ɹɾػց຋༁ XFBSFIJSJOH ɹɾը૾ೝࣝʁ લ൒ͷ·ͱΊ IUUQSFDSVJUDPPLQBEDPN

  26. ֶज़తͳݚڀΛࢧԉ͢ΔऔΓ૊Έ

  27. Α͋͘Δ૬ஊ ʮϨγϐͷσʔλΛ࢖͍͍ͨʂݚڀ͕ਐ·ͳ͍ʂʂʂʯ

  28. ૬ஊ͝ͱʹఏڙɻ͚Ͳ ɾݸผରԠ͕େม ɹɾඞཁͳσʔλͷฉ͖औΓ ɹɾར༻ܖ໿ͷࡦఆɾక݁ ɹɾσʔλͷ४උɾૹ෇ ɾྉཧʹؔ͢Δݚڀ͕޿͕Βͳ͍ ɹɾ૬ஊ͕͋Ε͹σʔλΛఏڙ͍ͯ͠Δ͜ͱΛ୭΋஌Βͳ͍ ɹɾ݁ՌɺϓϩμΫτʹ׆༻Ͱ͖Δݚڀ੒Ռ͕ࣦΘΕΔ

  29. σʔλެ։

  30. ɾݚڀऀ ɹɾϨγϐσʔλ͕ར༻Մೳ ɹɾڞ௨ͷσʔληοτΛ௨ͯ͠ɺઌߦݚڀͱͷൺֱ͕༰қʹ ɾΫοΫύου ɹɾݸผରԠͷख͕ؒݮগʢσʔλɾར༻ܖ໿ͷҰຊԽʣ ɹɾྉཧʹؔ͢Δݚڀ͕૿Ճ 8JO8JO

  31. ެ։σʔλʢϨγϐʣ ɾλΠτϧ ɾ֓ཁ ɾࡐྉ ɾखॱ ɾίπɾϙΠϯτ ɾ͜ͷϨγϐͷੜཱ͍ͪ ɾͭ͘ΕΆ ɾFUD

  32. ެ։σʔλʢݙཱʣ ɾλΠτϧ ɾ͜ͷݙཱͷϙΠϯτ ɾௐཧ࣌ؒ ɾஈऔΓͷίπ ɾओࡊ ɾ෭ࡊ ɾࢀߟʹͳͬͨʂ ɾ͜ͷݙཱͷओͳ৯ࡐδϟϯϧγʔϯ

  33. ར༻ํ๏ ɾ/**Λ௨ͯ͡ར༻Մೳ ɹɾIUUQXXXOJJBDKQDTDFOUFSJESDPPLQBEDPPLQBEIUNM ɾެతͳݚڀػؔͷํ͕ར༻Մೳ ɹɾاۀͷݚڀػؔͷํ͸ผ్͝૬ஊ͍ͩ͘͞ ɾར༻ن໿͸ҎԼͷϖʔδ ɹɾIUUQTDPPLQBEDPNUFSNTDPPLQBE@EBUB

  34. ਃ੥खॱ

  35. ਃ੥खॱ

  36. ίϯτϦϏϡʔγϣϯ ɾެ։͔Β࢛ϲ݄Ͱݚڀࣨʹఏڙ ɹɾجຊతʹ͸େֶ͔Βͷਃ੥ʢͨ·ʹߴઐͳͲ͔Β΋ʣ ɾاۀ͔Β΋ෳ਺ͷਃ੥ ɹɾఏڙՄೳ͔͸Ҋ݅࣍ୈʢݚڀ໨తͰͳ͍Ҋ݅͸μϝʣ ɾࠓޙɺڞ௨ͷσʔληοτͱͯ͠ར༻Մೳ ɹɾઌߦݚڀͱͷൺֱ͕༰қʹ

  37. ϋοΧιϯʢ༨ஊʣ ४උͷख͕ؒল͚ͨ

  38. ɾֶज़తͳݚڀΛࢧԉ͢ΔऔΓ૊Έ ɹɾΫοΫύουσʔλͷެ։ ɹɾݚڀࣨʹఏڙࡁΈ ɾࠓޙͷల๬ ɹɾը૾ެ։ʢ༧ఆʣ ɹɾ৯ϝσΟΞݚڀձͷ։࠵ʢ݄ʣ ޙ൒ͷ·ͱΊ

  39. ·ͱΊ

  40. ·ͱΊ ຊൃදͰ͸ɺΫοΫύουʹ͓͚Δ࠷৽ͷݚڀͷ׆༻ࣄྫͱֶज़తͳݚڀΛ ࢧԉ͢ΔͨΊͷऔΓ૊Έʹ͍ͭͯ঺հ͠·ͨ͠ɻΫοΫύουͰ͸ɺࠓ݄ɺ ݚڀྖҬΛઐ໳ʹѻ͏νʔϜ΋ൃ଍͠·ͨ͠ɻࠓޙɺΑΓҰ૚ݚڀྖҬʹྗ ΛೖΕ͍ͯ͘༧ఆͰ͢ɻ ΫοΫύου ݚڀ