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

会社訪問アプリ Wantedly Visit における 相互推薦システムの活用事例

chimuichimu
May 27, 2024
490

会社訪問アプリ Wantedly Visit における 相互推薦システムの活用事例

chimuichimu

May 27, 2024
Tweet

Transcript

  1. © 2024 Wantedly, Inc. ձࣾ๚໰ΞϓϦ Wantedly Visit ʹ͓͚Δ ૬ޓਪનγεςϜͷ׆༻ࣄྫ May.

    28 2024 - ࢢଜઍߊʢ΢ΥϯςουϦʔגࣜձࣾʣ JSAI 2024 ΠϯμετϦΞϧηογϣϯ
  2. © 2024 Wantedly, Inc. ໊લɿ ࢢଜ ઍߊ (Ichimura Chiaki) ॴଐͱ໾ׂɿ

    ɾ΢ΥϯςουϦʔגࣜձࣾ (2024/3~) ɾσʔλαΠΤϯςΟετ ɾਪનγεςϜͷ։ൃɾӡ༻ʹैࣄ @chimuichimu1 ࣗݾ঺հ
  3. © 2024 Wantedly, Inc. iOS, Android and Web ؾܰʹձࣾ๚໰ ϛογϣϯ΍Ձ஋؍΁ͷڞײͰϚονϯά

    څ༩΍෱རްੜͳͲͷ৚݅Ͱ͸ͳ͘ɺ૝͍͕͋Ε͹ձࣾͷن ໛ʹͱΒΘΕͳ͍ ·ͣʮ࿩Λฉ͖ʹߦ͘ʯͱ͍͏৽͍͠ମݧ ݸਓͱاۀ͕ϑϥοτͳ໨ઢͰग़ձ͑Δ͜ͱͰɺΑΓັྗత ͳ৔ॴΛݟ͚ͭΔ͜ͱ͕Մೳʹ ձࣾ๚໰ΞϓϦʮWantedly Visitʯ
  4. © 2024 Wantedly, Inc. Ϛονϯάͷਪનͷ೉͠͞ᶃ ૒ํ޲ͷᅂ޷͕ຬͨ͞Εͳ͚Ε͹Ϛονϯά͸੒ཱ͠ͳ͍ ڵຯ͋Γ ੒ޭ ࣦഊ Ұൠతͳਪન

    (Item to User) Ϛονϯάͷਪન (User to User) ੒ޭ ڵຯ͋Γ ڵຯ͋Γ ڵຯ͋Γ ڵຯͳ͠ Ұํ޲ͷᅂ޷͚ͩͰͳ͘ ૒ํ޲ͷᅂ޷ʹج͍ͮͨਪન͕ඞཁ
  5. © 2024 Wantedly, Inc. Ϛονϯάͷਪનͷ೉͠͞ᶄ ภͬͨਪનʹΑΔϢʔβʔମݧͷѱԽ Ұൠతͳਪન (Item to User)

    ※ ͨͩ͠ଟ༷ੑͷܽ೗΍ίϯςϯπੜ࢈ऀࢹ఺Ͱͷࡏݿ΍ެฏੑ͕໰୊ʹͳΔ৔໘͕͋Δ ͜ΜͳʹϝοηʔδΛ΋Βͬͯ΋ ରԠͰ͖ͳ͍… શવϝοηʔδΛ΋Β͑ͳ͍… Ϛονϯάͷਪન (User to User) ຊ౰ʹྑ͍΋ͷ͕ਪનͰ͖͍ͯΕ͹ ϢʔβʔମݧͷѱԽʹܨ͕Γʹ͍͘ʢ*ʣ ΩϟύγςΟΛ௒͑Δ਺ͷ޷ҙ΍ Ϛονϯά͕શ͘Ͱ͖ͳ͍ͱ͍͏ϢʔβʔମݧͷѱԽʹ௚݁
  6. © 2024 Wantedly, Inc. Ϛονϯάͷਪનͷ೉͠͞ᶅ ϢʔβʔͷߦಈಛੑʹىҼ͢Δ೉͠͞ Ұൠతͳਪન (Item to User)

    Ϣʔβʔ͸ࣗ਎ͷ໨తΛୡ੒͢ΔͨΊ ೳಈతʹߦಈΛߦ͏ ೳಈతʹߦಈ͢ΔϢʔβʔ΋͍Ε͹ ૬ख͔Β޷ҙ͕དྷΔͷΛ଴ͭडಈతͳϢʔβʔ΋͍Δ Ϛονϯάͷਪન (User to User) ❌ डಈతͳϢʔβʔ͸ద੾ͳϢʔβʔʹਪન͞Εͳ͍ݶΓ Ϛονϯάͱ͍͏໨త͕ୡ੒Ͱ͖ͳ͍
  7. © 2024 Wantedly, Inc. ૬ޓਪનγεςϜʢReciprocal Recommender Systemsʣͱ͸ʁ αʔϏε಺ͷϢʔβʔΛޓ͍ʹਪન͠߹͏γεςϜ • ҰൠతͳਪનγεςϜʢItem

    to User ͷਪનʣ • Ϣʔβʔ͔ΒΞΠςϜ΁ͷᅂ޷ʹج͍ͮͯਪન • ૬ޓਪનγεςϜʢUser to User ͷਪનʣ • ਪન͞ΕΔϢʔβʔ / ਪનΛड͚ΔϢʔβʔͷ྆ํͷᅂ޷ʹ ج͍ͮͯਪન
  8. © 2024 Wantedly, Inc. ૬ޓਪનγεςϜͷجຊతͳΞʔΩςΫνϟ User A to User B

    Preference Score User B to User A Preference Score Reciprocal Preference Score Aggregation Function Ϣʔβʔ A ͔ΒϢʔβʔ B ΁ͷᅂ޷είΞ Ϣʔβʔ B ͔ΒϢʔβʔ A ΁ͷᅂ޷είΞ ΛͦΕͧΕਪ࿦ ૬ޓͷᅂ޷είΞΛௐ࿨ฏۉͳͲͷؔ਺Ͱू໿ ྆ํͱ΋είΞ͕ߴ͍ͱߴ͘ɺͲͪΒ͔͕௿͍ ͱ௿͘ͳΔੑ࣭ͷؔ਺͕ద੾ͱݴΘΕ͍ͯΔ ू໿͞ΕͨείΞ͸૬ޓͷᅂ޷ͷϚον౓Λදݱ ͜Εʹج͍ͮͯਪનΛߦ͏
  9. © 2024 Wantedly, Inc. Wantedly Visit ΁ͷ૬ޓਪનγεςϜͷಋೖ Ϣʔβʔ اۀ Ϣʔβʔ

    → اۀͷᅂ޷༧ଌ اۀ → Ϣʔβʔͷᅂ޷༧ଌ ֤Ϟσϧͷ༧ଌ஋Λू໿ Ϛον౓͕ߴ͍ϢʔβʔΛ اۀʹਪન Ϛον౓͕ߴ͍ืूΛ Ϣʔβʔʹਪન
  10. © 2024 Wantedly, Inc. Wantedly Visit ΁ͷ૬ޓਪનγεςϜͷಋೖ Ϣʔβʔ اۀ Ϣʔβʔ

    → اۀͷᅂ޷༧ଌ اۀ → Ϣʔβʔͷᅂ޷༧ଌ ֤Ϟσϧͷ༧ଌ஋Λू໿ Ϛον౓͕ߴ͍ϢʔβʔΛ اۀʹਪન Ϛον౓͕ߴ͍ืूΛ Ϣʔβʔʹਪન Wantedly Visit ͷσʔλ͔Βਪનީิ ͱಛ௃ྔΛநग़ ɾϢʔβʔ / اۀͷߦಈϩά ɾϢʔβʔ / اۀͷଐੑ ɾϓϩϑΟʔϧͳͲͷςΩετ৘ใ
  11. © 2024 Wantedly, Inc. Wantedly Visit ΁ͷ૬ޓਪનγεςϜͷಋೖ Ϣʔβʔ اۀ Ϣʔβʔ

    → اۀͷᅂ޷༧ଌ اۀ → Ϣʔβʔͷᅂ޷༧ଌ ֤Ϟσϧͷ༧ଌ஋Λू໿ Ϛον౓͕ߴ͍ϢʔβʔΛ اۀʹਪન Ϛον౓͕ߴ͍ืूΛ Ϣʔβʔʹਪન ػցֶशϞσϧͰ֤ํ޲ͷᅂ޷Λ༧ଌ ɾϢʔβʔ͕ͦͷืूʹڵຯΛ͔࣋ͭʁ ɾاۀ͕ͦͷϢʔβʔʹڵຯΛ͔࣋ͭʁ
  12. © 2024 Wantedly, Inc. Wantedly Visit ΁ͷ૬ޓਪનγεςϜͷಋೖ Ϣʔβʔ اۀ Ϣʔβʔ

    → اۀͷᅂ޷༧ଌ اۀ → Ϣʔβʔͷᅂ޷༧ଌ ֤Ϟσϧͷ༧ଌ஋Λू໿ Ϛον౓͕ߴ͍ϢʔβʔΛ اۀʹਪન Ϛον౓͕ߴ͍ืूΛ Ϣʔβʔʹਪન ͦΕͧΕͷ༧ଌ஋ΛॏΈ͚ͮͯ͠ू໿ ૒ํ޲ͷᅂ޷ʹجͮ͘ʮϚον౓ʯ͕ ಘΒΕΔ
  13. © 2024 Wantedly, Inc. Wantedly Visit ΁ͷ૬ޓਪનγεςϜͷಋೖ Ϣʔβʔ اۀ Ϣʔβʔ

    → اۀͷᅂ޷༧ଌ اۀ → Ϣʔβʔͷᅂ޷༧ଌ ֤Ϟσϧͷ༧ଌ஋Λू໿ Ϛον౓͕ߴ͍ϢʔβʔΛ اۀʹਪન Ϛον౓͕ߴ͍ืूΛ Ϣʔβʔʹਪન Ϛον౓Λ༻͍ͯ ɾϢʔβʔʹ͸Ԡืର৅ͷืूΛ ɾاۀʹ͸εΧ΢τର৅ͷީิऀΛ ͦΕͧΕਪન Ϛον౓͸૒ํͷᅂ޷ʹجͮͨ͘Ίɺ ᅂ޷ͷ͢Εҧ͍Λ཈ࢭ →ʮ྆ࢥ͍ʯͷਪનΛ࣮ݱ
  14. © 2024 Wantedly, Inc. ૬ޓਪનγεςϜͷಋೖʹΑΔ੒Ռ ԠืɾεΧ΢τͷ྆ํͰओཁ KPI ͷվળ 🎉 •

    ैདྷͷϕʔεϥΠϯʢItem to User ͷਪનʣͱൺֱͯ͠ • ϢʔβʔͷԠืىҼͷϚονϯά਺ɾޮ཰͕େ͖͘վળ • اۀͷεΧ΢τىҼͷϚονϯά਺ɾޮ཰͕େ͖͘վળ
  15. © 2024 Wantedly, Inc. ղܾ͢΂͖ٕज़త՝୊ᶄ Ϣʔβʔͷߦಈಛੑʹج͍ͮͨϞσϦϯά Ͳ͏ղܾʹ޲͔͏͔ʁ • डಈతͳϢʔβʔ͕ద੾ͳೳಈతͳϢʔβʔʹਪન͞Ε΍͘͢ͳΔΑ͏ ͳਪનΞʔΩςΫνϟͷݕ౼

    • ϓϩϑΟʔϧ౳ͷίϯςϯπΛ׆༻͠डಈతͳϢʔβʔͷᅂ޷Λଊ͑Δ ؔ࿈ݚڀ • Ϣʔβʔͷ Active / Passive ͳߦಈͷάϥϑදݱͱ BERT ʹΑΔςΩε τͷຒΊࠐΈ͔ΒϚονϯάΛߦ͏ݚڀ [Yang et al., 2022]
  16. © 2024 Wantedly, Inc. ·ͱΊ • ΢ΥϯςουϦʔ͸ʮγΰτͰίίϩΦυϧͻͱʯΛ૿΍ͨ͢Ίʹɺਪ નٕज़Λ׆༻ͨ͠ϚονϯάͷମݧΛఏڙ͍ͯ͠Δ • ϚονϯάͷਪનͰ͸ɺϢʔβʔͱاۀͷ૒ํͷᅂ޷Λຬͨ͢͜ͱ͕ॏ

    ཁɻ૬ޓਪનγεςϜΛಋೖ͢͠Δ͜ͱͰʮ྆ࢥ͍ʯͷਪનΛ࣮ݱ͠ ϢʔβʔͱاۀͷϚονϯάମݧΛେ͖͘վળͨ͠ • ϢʔβʔͷΩϟύγςΟ΍ߦಈಛੑΛߟྀͨ͠ਪનͷ࣮ݱͳͲɺղܾ͢ ΂͖ٕज़త՝୊͸·ͩ·ͩ͋Δ
  17. © 2024 Wantedly, Inc. ࢀߟจݙ • Yi Su, Magd Bayoumi,

    and Thorsten Joachims. 2022. Optimizing Rankings for Recommendation in Matching Markets. In Proceedings of the ACM Web Conference 2022 (WWW '22). Association for Computing Machinery, New York, NY, USA, 328–338. https://doi.org/10.1145/3485447.3511961 • Chen Yang, Yupeng Hou, Yang Song, Tao Zhang, Ji-Rong Wen, and Wayne Xin Zhao. 2022. Modeling Two- Way Selection Preference for Person-Job Fit. In Proceedings of the 16th ACM Conference on Recommender Systems (RecSys '22). Association for Computing Machinery, New York, NY, USA, 102–112. https://doi.org/ 10.1145/3523227.3546752