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

コーディネート整合性を考慮したカテゴリ間推薦

18d356eace1ae9fb3c1188871990a482?s=47 Hayato Maki
July 04, 2018
0
1.1k

 コーディネート整合性を考慮したカテゴリ間推薦

presented in CCSE at Meiji University on 2018.07.01

18d356eace1ae9fb3c1188871990a482?s=128

Hayato Maki

July 04, 2018
Tweet

More Decks by Hayato Maki

See All by Hayato Maki
Billion-scale Embedding for E-commerce Recommendation in Alibaba
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
0
18
Today was a Good Day: The Daily Life of Software Developers
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
0
8
論文紹介：Relaxed Softmax for PU Learning
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
3
700
MIRU 2019 Lunch on Seminar
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
1
110
Regularization_The Element of Statical Learning
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
0
110
Neural Activity During Sentence Processing as Reflected in Theta, Alpha, Beta, and Gamma Oscillations
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
0
77
【ICML読み会】Unsupervised Deep Embedding for Clustering Analysis
18d356eace1ae9fb3c1188871990a482?s=48 hamaki
0
1.1k

Featured

See All Featured
How GitHub (no longer) Works
78b475797a14c84799063c7cd073962f?s=48 holman
297
140k
10 Git Anti Patterns You Should be Aware of
Dafc4723e9a1c067796c0fec6f509774?s=48 lemiorhan
638
53k
How GitHub Uses GitHub to Build GitHub
78b475797a14c84799063c7cd073962f?s=48 holman
465
280k
Docker and Python
Ecdea9b9714877b86cee08458f085481?s=48 trallard
27
1.6k
Art, The Web, and Tiny UX
D67fff74178013024d833b3eec6cf27f?s=48 lynnandtonic
280
18k
Java REST API Framework Comparison - PWX 2021
72a2082c6a4dd79ad68befb3db911616?s=48 mraible
PRO
11
4.9k
Making the Leap to Tech Lead
32de0bd2ba869609d26fd052a4622778?s=48 cromwellryan
113
7.4k
Support Driven Design
1519587b1a0febb658c36c94f6272b0d?s=48 roundedbygravity
87
8.6k
Facilitating Awesome Meetings
245cee81a9c424266e5e401d844ea881?s=48 lara
29
4.1k
A designer walks into a library…
15f2b8221f247985a27a627d2ece72f0?s=48 pauljervisheath
197
16k
Unsuck your backbone
B83d5b590577969ee79a5ad845411a7d?s=48 ammeep
659
55k
Agile that works and the tools we love
462dad0dd24c568a32dcdb0ae895ae1b?s=48 rasmusluckow
319
19k

Transcript

  1. ίʔσΟωʔτ੔߹ੑΛ
 ߟྀͨ͠ΧςΰϦؒਪન ਅ໦༐ਓ
 ελʔττΡσΠݚڀॴ presented in CCSE at ໌࣏େֶ on

    2018.07.01

  2. ϑΝογϣϯʹ͓͚ΔਪનγεςϜ 2 өը Իָ χϡʔε ॻ੶ ϑΝογϣϯ ΞΠςϜ͸̍ͭͣͭݸผʹར༻͞ΕΔ ΞΠςϜ͸ෳ਺ಉ࣌ʹར༻͞ΕΔ 

    ΞΠςϜಉ࢜ͷ૬ੑͷ໰୊͕ൃੜ ෰ͷʮ૊Έ߹Θͤͷྑ͞ʯΛ
 ίʔσΟωʔτͷ੔߹ੑ DPNQBUJCJMJUZ ͱݺͿ

  3. ΧςΰϦؒͷਪન 3 ΧςΰϦ" ͷΞΠςϜ ΧςΰϦ# ͷΞΠςϜ ΧςΰϦؒਪન
 γεςϜ ೖྗ ग़ྗ

    ྫ τοϓε ྫ ϘτϜε ίʔσΟωʔτͷ੔߹ੑ͕ߴ͍ਪનΛ͍ͨ͠

  4. ઌߦݚڀ

  5. 5 ΧςΰϦؒਪનͷ̎εςοϓ ಛ௃நग़εςοϓ ಛ௃நग़ث ಛ௃நग़ث ۙ๣୳ࡧεςοϓ ڞ௨ͷۭؒ΁ ຒΊࠐΉ ಛ௃ϕΫτϧ ಛ௃நग़ث

    ಛ௃ϕΫτϧ ೖྗΫΤϦ Ұ൪͍ۙ
 ΞΠςϜΛग़ྗ ΞΠςϜը૾

  6. 4JBNFTF$//ʹΑΔಛ௃நग़ 6 4JBNFTF$//ύϥϝλΛڞ༗ͨ̎ͭ͠ͷ$// $// $// ಛ௃ϕΫτϧ 4JBNFTF$// $POUSBTUJWF
 ଛࣦ $POUSBTUJWFଛࣦ

    [Bromley+, ’94] [Veit+, ’15] ೖྗ͕ਖ਼ྫͷ৔߹ɺϖΞ͕ۙͮ͘Α͏ʹֶश ෛྫͷ৔߹ɺϖΞ͕ԕ͔͟ΔΑ͏ʹֶश xi xj ∥xi − xj ∥2 max(0,m − ∥xi − xj ∥2 ) m : ϋΠύʔύϥϝλ

  7. .JYUVSFPG&YQFSUʹΑΔॊೈͳਪન 7 ίʔσΟωʔτʹ͸ෳ਺ͷʮྑ͞ʯ͕͋Δ  ෳ਺ͷϞσϧΛֶश .JYUVSFPG&YQFSUT ϚϋϥϊϏεڑ཭ [He+, ’16] xi

    xj γάϞΠυ
 ؔ਺ (xi xj)> M (xi xj) <latexit sha1_base64="Tl21jfMN4QyHTFwXcefcHVg0v3A=">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</latexit> <latexit sha1_base64="Tl21jfMN4QyHTFwXcefcHVg0v3A=">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</latexit> <latexit sha1_base64="Tl21jfMN4QyHTFwXcefcHVg0v3A=">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</latexit> <latexit sha1_base64="Tl21jfMN4QyHTFwXcefcHVg0v3A=">AAAEh3iclVPLbtNAFL0phpbwaAobJDYRUVFZEE1QVUpXBTZskPogbaU6RLYZO9P6JXuStljhA/oDLFiBxALxF7DhB1j0D0CsoEhsWHBm7PCqCOlYtu899577mDtjx75IJWMHpbETxslT4xOny2fOnjs/WZm6sJZG3cThTSfyo2TDtlLui5A3pZA+34gTbgW2z9ft7bvKvt7jSSqi8IHci3krsLxQuMKxJKB2xTR97soZM7Bkx3az3X5bXP9N2TIT4XXktYemjOLqwHC/PxqrXamxOtOrelRoFEKNirUUTZX2yaRHFJFDXQqIU0gSsk8WpXg2qUGMYmAtyoAlkIS2c+pTGdwuvDg8LKDb+HrQNgs0hK5ipprtIIuPNwGzStPsPXvFDtk79pp9ZN//GSvTMVQte/jbOZfH7cn9S6vf/ssK8JfU+cUaUvPANwWa6Iq59lSa0HswvGNJLs1rnkDnsUbUHjh59t7jp4erCyvT2VX2gn1C98/ZAXuL/sPeV+flMl95NqQbG9Xl2UPoO3ovA11xiOllwNVsPCAqd18jg34iRFN6AqRa+D356WliJkJrEplHy2HR7og5cs/j5Aj1CVHdqrPX0pg6bS64SlqDrE5hDVZTT8ZDfqnlgW3YnEaaMu5R4+9bc1Ro3qjfqjeWZ2uLd4oLNUGX6QrNoLSbtEj3aImaCPyGPtBn+mKUDWbMGfO561ip4FykP5Zx+weDogra</latexit> ϚϋϥϊϏεڑ཭ ϚϋϥϊϏεڑ཭ ιϑτϚοΫε ɾ
 ɾ ɾ ಛ௃ϕΫτϧ

  8. ໰୊఺ 8 w ख๏ʹؔ͢Δվળ͸৭ʑɾɾɾ w σʔληοτͷ໰୊  "NB[POಉ࣌ߪೖσʔληοτ  ಉ࣌ʹߪೖ͞ΕͨʹίʔσΟωʔτ੔߹ੑ͕ߴ͍ͱΈͳ͢

    "NB[POಉ࣌ߪೖσʔλͷ໰୊఺ w ಉ࣌ߪೖ͞ΕͨΞΠςϜίʔσΟωʔτͱͯ͠ ಉ࣌ண༻͞ΕΔΞΠςϜ w ͓͠ΌΕʹؔ৺ͷແ͍Ϣʔβͷํ͕ଟ͍ w ߪೖͨ͠ΞΠςϜண༻͍ͨ͠ΞΠςϜ

  9. ϑΝογϣϯ4/4ͷσʔλ 9 w*20/  ສϢʔβͷϑΝογϣϯ4/4  Ϣʔβ͕෰ͷը૾Λ૊Έ߹ΘͤͯίʔσΟωʔ τΛ࡞੒ɺ౤ߘ ౤ߘͷྫ

  10. *20/σʔλͷར఺ 10 w ʢಉ࣌ߪೖ౳Ͱ͸ͳ͘ʣίʔσΟωʔτͱ ͯ͠ಉ࣌ண༻Λ૝ఆͨ͠ը૾ηοτ w ΦγϟϨʹؔ৺ͷ͋ΔϢʔβʔ͕࡞੒ w ʮങͬͨ෰ʯͰ͸ͳ͘ʮண͍ͨ෰ʯΛ൓ө

  11. ࣮ɹݧ

  12. ࣮ݧઃఆ 12 ಛ௃நग़ωοτϫʔΫ *ODFQUJPO7 ଛࣦؔ਺ $POUSBTUJWFଛࣦ ॳظ஋ ֶशࡁΈ෺ମೝࣝϞσϧ ݕࡧʹ༻͍Δ
 ಛ௃ϕΫτϧͷ࣍ݩ

     ਖ਼ྫσʔλ *20/ͰίʔσΟωʔτͱͯ͠౤ߘ ͞ΕͨΞΠςϜϖΞ ෛྫσʔλ *20/ͰίʔσΟωʔτͱͯ͠౤ߘ ͞Εͳ͔ͬͨΞΠςϜϖΞ

  13. 13 ݁Ռ ֶशલ ΤϙοΫ
 ֶशޙ ֶशલ ΤϙοΫ
 ֶशޙ τοϓεUPϘτϜε

  14. 14 ݁Ռ ֶशલ ΤϙοΫ
 ֶशޙ ֶशલ ΤϙοΫ
 ֶशޙ τοϓεUPόοά ֶशલͷϞσϧ͸෺ମೝࣝͰ܇࿅͞Ε͍ͯΔͨΊɺݟͨ໨͕ࣅ͍ͯΔΞΠςϜ

    Λਪન͢Δ܏޲ʹ͋Δ

  15. ·ͱΊ 15 ໨త ίʔσΟωʔτ੔߹ੑͷߴ͍ΧςΰϦؒਪન ख๏ $//ʹΑΔಛ௃நग़ͱ$POUSBTUJWFଛࣦʹΑΔɺ ϖΞը૾ͷڞ௨ۭؒ΁ͷຒΊࠐΈ ݁Ռ ༧උ࣮ݧͷ݁ՌɺֶशʹΑͬͯίʔσΟωʔτ੔ ߹ੑΛֶशՄೳͳ͜ͱ͕෼͔ͬͨɻ

    ݱࡏɺϞσϧͷ࡮৽ʹऔΓ૊ΜͰ͍Δɻ