The Web Conference2020 参加報告会

Slide 1

Slide 1 text

Slide 2

Slide 2 text

Slide 3

Slide 3 text

©2020 Wantedly, Inc. 1. Deep Transfer Learning for Search and Recommendation ɾݕࡧɾਪનγεςϜʹ͓͚ΔసҠֶशͷԠ༻ͱLinkedinͰͷࣄྫΛ঺հ͍ͯ͠Δ (Tutorial) 2. Clustering and Constructing User Coresets to Accelerate Large- scale Top-K Recommender Systems ɾਪનର৅ͷϢʔβΛᅂ޷ͷ͍ۙ΋ͷಉ࢜ͰΫϥελϦϯά͠, ֤Ϋϥελͷ୅දϕΫτϧΛ  ɹ ࢖ͬͯۙࣅ࠷ۙ๣୳ࡧΛ͢Δ͜ͱͰਪનγεςϜͷεϐʔυΞοϓΛ࣮ݱ 3. Adaptive Hierarchical Translation-based Sequential Recommendation ɾϢʔβߦಈͷ࣌ܥྻΛߟྀͨ͠Translation-based Reccommendationʹ, ΞΠςϜؒͷ  ɹ ؔ܎ੑͱ࣌ؒґଘͷuser translation vectorΛՃ͑ͨख๏ΛఏҊ ຊൃදͰ঺հ͢Δ࿦จɾηογϣϯ

Slide 4

Slide 4 text

Slide 5

Slide 5 text

Slide 6

Slide 6 text

©2020 Wantedly, Inc. Deep Transfer Learning 1. Model Transfer ɹɾsource domainsͰͷֶशࡁΈϞσϧΛtarget domainͰ࢖͏ ɹɾfine-tuning (e.g. BERT) ΍ ֤domainͰͷtaskͷmulti-task learning 2. Feature Representation Transfer ɹɾҟͳΔdomainؒͰڞ༗Մೳͳಛ௃Λֶश ɹɾsource domainͰͷֶशࡁΈϞσϧ͔Βͷಛ௃நग़ ΍ domain adaptation 3. Instance Transfer ɹɾsource domain͔Β܇࿅σʔλΛ࣋ͬͯ͘Δ

Slide 7

Slide 7 text

©2020 Wantedly, Inc. Deep Transfer Learning for Search and Recommendation ݕࡧɾਪનγεςϜͷͲͷ෦෼ͰసҠֶशΛ࢖͏͔ Raw Query Query Understanding User Profile, Activities Raw Document User Understanding Document Understanding Document Retrieval Ranking ←ΑΓྑ͍ཧղͷͨΊʹ, pretrained model  ͔ΒຒΊࠐΈදݱΛྲྀ༻ (Feature Transfer) ←ຒΊࠐΈۭؒ಺ͰͷީิΞΠςϜͷ֦ு (Feature Transfer) ←ؔ࿈͢ΔαʔϏεͷσʔλ΍ϞσϧΛ  ྲྀ༻͢Δ(Model Transfer & Instance Transfer)

Slide 8

Slide 8 text

Slide 9

Slide 9 text

©2020 Wantedly, Inc. Case study in Linkedin (2) People Search + Careers Job Recommendation ɾ֤υϝΠϯͷਪનλεΫΛղ͍ͯuser, itemͷදݱΛֶश͠, ಛ௃ྔͱͯ͠ར༻ Add embeddings trained with... AUC Lift NDCG@25 Lift Tree Model Baseline  (no embedding added) 0.00% 0.00% + People Search Embedding (Single-task) 1.25% 0.85% + Multi-task Embedding 1.82% 1.50%

Slide 10

Slide 10 text

Slide 11

Slide 11 text

©2020 Wantedly, Inc. Motivation ྨࣅΞΠςϜͷ୳ࡧͷߴ଎Խ ɾϢʔβʹਪન͢ΔΞΠςϜΛ୳͢ࡍ, PredictionʹؔΘΔ࣮ߦ࣌ؒΛશମతʹ୹͍ͨ͘͠ ɹɾPrediction time = Preparation time + Inference time ɹɾطଘͷ୳ࡧΞϧΰϦζϜͰ͸Inferece timeͷ୹ॖʹয఺Λ౰͍ͯͯΔ  ɾطଘͷख๏Ͱ͸Ϣʔβؒͷؔ܎ੑΛߟྀ͍ͯ͠ͳ͍ͷͰ, ߴ଎Խʹ׆༻͍ͨ͠ ɹɾ࠷ۙ๣୳ࡧ͸೚ҙͷΫΤϦϕΫτϧ͕༩͑ΒΕͨ࣌ʹ͍ۙϕΫτϧΛ୳͢໰୊͕ͩ,   ɹɹਪનγεςϜͰ͸ΫΤϦϕΫτϧ͸ϢʔβϕΫτϧͰ΋͋Γ, ϢʔβϕΫτϧ͸ڧ͍  ɹɹΫϥελϦϯάߏ଄Λ͍࣋ͬͯΔ ɹɾ͍ۙϢʔβ͸ࣅͨؔ৺Λ͍࣋ͬͯΔͨΊ, ୳ࡧ࣌ʹ·ͱΊͯ͠·͑͹ޮ཰తͰ͸?

Slide 12

Slide 12 text

©2020 Wantedly, Inc. Framework Overview Preparation Stage: 1. αϒαϯϓϧͨ͠ϢʔβΛΫϥελϦϯά 2. άϧʔϓ಺ͷϢʔβΛΧόʔͰ͖ΔΑ͏ͳ  ෳ਺ͷ୅දϕΫτϧΛϢʔβάϧʔϓຖʹٻΊΔ 3. ΞΠςϜͷ֊૚ܕάϥϑߏ଄Λߏங͠, ɹ֤Ϣʔβάϧʔϓͷ୅දϕΫτϧΛ༻͍ͨ  ɹgreedyͳάϥϑ୳ࡧʹΑΓ, άϧʔϓຖʹ  ɹީิΞΠςϜू߹Λऔಘ Prediction Stage: 4. ༧ଌର৅Ϣʔβͷॴଐ͢ΔάϧʔϓΛ୳͠, ɹͦͷάϧʔϓʹରԠ͢ΔީิΞΠςϜू߹ͱ ɹϢʔβϕΫτϧͷ಺ੵΛͱͬͯϥϯΫ෇͚

Slide 13

Slide 13 text

Slide 14

Slide 14 text

Slide 15

Slide 15 text

Slide 16

Slide 16 text

Slide 17

Slide 17 text

Slide 18

Slide 18 text

©2020 Wantedly, Inc. Motivation ΞΠςϜؒͷؔ࿈ੑͱϢʔβͷߦಈܥྻΛߟྀͨ͠ਪનΛ͍ͨ͠ ɾϢʔβͷߦಈܥྻʹج͍ͮͨਪન (Sequential recommendation) ɹɾe.g. RNN, Ϛϧίϑ࿈࠯, CNN, Translated-based Recommendation ɾϢʔβͷߪങܥྻͷཧղʹ͸ΞΠςϜؒͷ૬ิؔ܎ͱ୅ସؔ܎͕ॏཁ ɹɾPCΛߪೖͨ͠௚ޙʹผͷPCΛങ͏͜ͱ͸ͳ͍͕, ͕࣌ؒܦͬͨΒങ͍׵͑ͨ͘ͳΔ