DEIM2023 大規模Webデータにおけるキーワード・地域ごとの拡散パターン抽出

Transcript

1. େن໛Webσʔλʹ͓͚Δ Ωʔϫʔυɾ஍Ҭ͝ͱͷ֦ࢄύλʔϯநग़ ଜࢁଠҰ, দݪ༃ࢠ, ᓎҪอࢤ େࡕେֶ, ࢈ۀՊֶݚڀॴ DEIM 2023
2. 2 ιʔγϟϧηϯαʔͷ໾ׂ എܠ ιʔγϟϧηϯαʔ ࣮ࡍͷࣾձ ౤ߘݕࡧ ϑΟʔυόοΫ
3. 3 Google Flu എܠ
4. 4 ݕࡧΫΤϦ͔Βજࡏతͳؔ܎Λଊ͑ΒΕΔ͔ʁ Ϟνϕʔγϣϯ U.S. Japan Italy Facebook Twitter Instagram Facebook Twitter Instagram Q: Ωʔϫʔυؒͷؔ܎͸ʁ Q: 国同⼠の関係は？ ？
5. 5 ໰୊ఆٛ (JWFO ΦϯϥΠϯ׆ಈσʔλ e.g., άʔάϧݕࡧ
6. 6 ໰୊ఆٛ (PBM
    ஍Ҭؒͷ֦ࢄύλʔϯ e.g., άʔάϧݕࡧ taylor swift lady gaga beyonce katy perry maroon 5 stevie wonder Group A: Russia, Mexico, Spain, Vietnam, etc. Group B: United Kingdom, Germany, France, etc. Group C: United States, Canada, Australia, etc. Group D: Iran, Pakistan, Brazil, Argentina, etc. Group E: China, Japan, Vietnam, South Africa, etc.
7. 7 ໰୊ఆٛ (PBM
    Ωʔϫʔυؒͷؔ܎ e.g., άʔάϧݕࡧ United States lady gaga maroon 5 stevie wonder beyonce katy perry taylor swift Japan lady gaga maroon 5 stevie wonder beyonce katy perry taylor swift
8. 8 ໰୊ఆٛ (PBM
    قઅੑ e.g., άʔάϧݕࡧ
9. 9 ໰୊ఆٛ (PBM
    ߴ͍༧ଌਫ਼౓ e.g., άʔάϧݕࡧ 2012 2014 2016 2018 Time Keywords Countries Russia Mexico Colombia Modeling Forecasting beyonce katy perry stevie wonder taylor swift lady gaga maroon 5
10. 10 ໰୊ఆٛ Given
    ΦϯϥΠϯ׆ಈσʔλ Goal 1 ஍Ҭؒͷ֦ࢄύλʔϯ Goal 2 Ωʔϫʔυؒͷؔ܎ Goal 3 قઅੑ Goal 4
    ߴ͍༧ଌਫ਼౓
11. 11 ໰୊ఆٛ Our solution: FLUXCUBE 拡散反応⽅程式 + ニューラルネットワーク Given
    ΦϯϥΠϯ׆ಈσʔλ Goal 1 ஍Ҭؒͷ֦ࢄύλʔϯ Goal 2 Ωʔϫʔυؒͷؔ܎ Goal 3 قઅੑ Goal 4
    ߴ͍༧ଌਫ਼౓
12. 12 ൓Ԡ֦ࢄํఔࣜ Our model: FLUXCUBE 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 Main Equation ൓Ԡ߲ ෼ࢄ߲ Reaction-Diffusion System is utilized is a mathematical model corresponding for physical phenomena.
13. 13 Our model: FLUXCUBE 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 Main Equation ෼ࢄ߲ Amazon BestBuy ebay Target ڝ߹ؔ܎ ૬ޓؔ܎ Macy’s Washington State Amazon Target ebay Craigslist قઅ߲ ϒϥοΫϑϥΠσʔ + × ൓Ԡ֦ࢄํఔࣜ ൓Ԡ߲
14. 14 Our model: FLUXCUBE ೖྗ: 𝝌 = 𝒙𝒕𝒊𝒋 λΠϜεςοϓ U
    ஍Ҭ J
    ΞΠςϜ K 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 反応拡散⽅程式 FLUXCUBE 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& ൓Ԡ߲ ֦ࢄ߲ قઅ߲
15. 15 Our model: FLUXCUBE 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 反応拡散⽅程式 FLUXCUBE 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& ൓Ԡ߲ Amazon BestBuy ebay Target Competitive relationship Mutualistic relationship Macy’s
16. 16 ൓Ԡ߲: ΞΠσΞ ΞΠςϜͷؔ܎͸δϟϯάϧͷಈ෺ͷؔ܎ͱྨࣅ (Matsubara, 2015) ൓Ԡ߲: FLUXCUBE Spider monkeys Capybaras Squirrel monkeys Macaws Fruits Nuts Grass Facebook Twitter Youtube Reddit Kids Teens Adults 類似
17. 17 ൓Ԡ߲: ΞΠσΞ ϩτΧɾϘϧςϥํఔࣜ ൓Ԡ߲: FLUXCUBE Facebook Twitter Youtube Reddit Kids Teens Adults 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏&
18. 18 ൓Ԡ߲: ΞΠσΞ ϩτΧɾϘϧςϥํఔࣜ ൓Ԡ߲: FLUXCUBE Facebook Twitter Youtube Reddit Kids Teens Adults 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏& 𝒄𝒋𝒋 : ΩʔϫʔυKͱΩʔϫʔυK` ؒͷ૬ޓ࡞༻ͷڧ͞
19. 19 ൓Ԡ߲: ΞΠσΞ ϩτΧɾϘϧςϥํఔࣜ Reaction term: FLUXCUBE Facebook Twitter Youtube Reddit Kids Teens Adults 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏&
20. 20 ൓Ԡ߲: ΞΠσΞ ϩτΧɾϘϧςϥํఔࣜ ൓Ԡ߲: FLUXCUBE Facebook Twitter Youtube Reddit Kids Teens Adults 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏&
21. 21 ൓Ԡ߲: ΞΠσΞ ϩτΧɾϘϧςϥํఔࣜ ൓Ԡ߲: FLUXCUBE Facebook Twitter Youtube Reddit Kids Teens Adults 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏&
22. 22 ൓Ԡ߲: ΞΠσΞ Lotka-Volterra Equation Λ൓Ԡ߲ͱͯ͠ར༻ ൓Ԡ߲: FLUXCUBE 𝒇 𝑥$%& |𝝌𝒕,𝒊,: = 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏& 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 FLUXCUBE:
23. 23 ֦ࢄ߲: FLUXCUBE 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 反応拡散⽅程式 FLUXCUBE 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& Seasonal term ֦ࢄ߲ Washington State Amazon Target ebay Craigslist
24. 24 ֦ࢄ߲: ΞΠσΞ l Ωʔϫʔυؒͷ૬ޓ࡞༻͸֎෦ཁҼͳͲʹΑΓ ҰఆͰ͸ͳ͍ l ෳࡶͳݱ৅΍ٸܹͳมԽͳͲʹΑͬͯੜ͡Δɼ ࣌ؒมԽ͢Δ૬ޓ࡞༻Λଊ͑Δඞཁ ֦ࢄ߲: FLUXCUBE Washington State Amazon Target ebay Craigslist
25. 25 ֦ࢄ߲: ΞΠσΞ l Ωʔϫʔυؒͷ૬ޓ࡞༻͸֎෦ཁҼͳͲʹΑΓ ҰఆͰ͸ͳ͍ l ෳࡶͳݱ৅΍ٸܹͳมԽͳͲʹΑͬͯੜ͡Δɼ ࣌ؒมԽ͢Δ૬ޓ࡞༻Λଊ͑Δඞཁ ֦ࢄ߲: FLUXCUBE Washington State Amazon Target ebay Craigslist 相互作⽤はPhysics-informed neural networksのアイデアを活⽤して RNNを利⽤
26. 26 Physics-informed neural Network (PINN) l ෺ཧϞσϧͷؔ਺͸ਂ૚ֶशͰۙࣅՄೳ l ॱ໰୊: ํఔࣜͷۙࣅղ l ٯ໰୊: ؍ଌσʔλ͔ΒύϥϝʔλΛਪఆ l ༩͑ΒΕֶͨशσʔλͱɺٻΊ͍ͨύϥϝʔλ͕ํఔࣜΛຬ ͨ͢Α͏ʹղۭؒΛ੍ݶ͢Δ͜ͱͰద੾ͳύϥϝʔλΛਪఆ ֦ࢄ߲: FLUXCUBE ภඍ෼ํఔࣜͰ͋Δ֦ࢄ൓Ԡํఔࣜͷ ύϥϝʔλͷҰ෦Λ/FVSBM
    /FUXPSL ͰॊೈʹදݱΛࢼΈΔ
27. 27 ֦ࢄ߲: FLUXCUBE ℊ 𝑥$%& |𝝌𝒕 , 𝒕 = 8 𝑅𝑁𝑁(1: 𝑡) ⊙ 𝝌𝒕 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 FLUXCUBE: l ૬ޓ࡞༻ͷਪఆʹRecurrent Neural NetworkΛར༻ l ॱ໰୊: ํఔࣜͷۙࣅղ l ٯ໰୊: ؍ଌσʔλ͔ΒύϥϝʔλΛਪఆ l ਺ཧϞσϧͷҰ෦ʹχϡʔϥϧωοτϫʔΫΛ׆༻͢Δ͜ͱ ͰॊೈͳϞσϦϯάͱߴ͍આ໌ੑΛཱ྆
28. 28 RNNͷग़ྗ஋ʹΑͬͯɺ֤Ωʔϫʔυͷ֦ࢄͷӨڹΛදݱ ֦ࢄ߲: FLUXCUBE ℊ 𝑥$%& |𝝌𝒕 , 𝒕 = 8 𝑅𝑁𝑁(1: 𝑡) ⊙ 𝝌𝒕 𝑹𝑵𝑵(𝟏: 𝒕) Time カナダからフランスへ の各キーワードの影響 の流れ
29. 29 قઅ߲: FLUXCUBE 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 拡散反応⽅程式 FLUXCUBE 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& قઅ߲ Washington State Amazon Target ebay Craigslist
30. 30 ΦϯϥΠϯσʔλ͸ΫϦεϚε΍ϒϥοΫϑϥΠσʔͳͲ ͷقઅΠϕϯτʹԠͯ͡มԽ قઅ੒෼ͷநग़ قઅ߲: FLUXCUBE 𝑥$)*%& = 𝐹 ,-!"# ,$ + 𝑥$%& = 1 + 𝑺$ ./0 1 %& ⊙ ,-!"# ,$ + 𝑥$%& 季節性の周期 e.g., p = 52週間 ϒϥοΫϑϥΠσʔ
31. 31 Our model: FLUXCUBE 𝜕𝒖 𝜕𝑡 = 𝒇 𝒖, 𝑡 + 𝒟Δ𝑢 反応拡散⽅程式 FLUXCUBE 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& ೖྗ: 𝝌 = 𝒙𝒕𝒊𝒋 λΠϜεςοϓ U
    ஍Ҭ J
    ΞΠςϜ K ൓Ԡ߲ ֦ࢄ߲ قઅ߲
32. 32 Fitting: FLUXCUBE Loss Function 𝝌𝒄 − @ 𝝌 𝟐 + 𝛼 8 𝐷 4 + 𝛽 8 𝑆 4 ճؼ߲ εύʔε߲ 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& λΠϜεςοϓ U
    ஍Ҭ J
    ΞΠςϜ K ೖྗ: 𝝌 = 𝒙𝒕𝒊𝒋
33. 33 l ଟ͘ͷ஍Ҭؒͷ૬ޓ࡞༻Λਪ࿦͢Δ͜ͱ͸ɺܭࢉίετ͕ߴ͍ ஍ҬͷάϧʔϓΠϯά
34. 34 l ଟ͘ͷ஍Ҭؒͷ૬ޓ࡞༻Λਪ࿦͢Δ͜ͱ͸ɺܭࢉίετ͕ߴ͍ ஍ҬͷάϧʔϓΫϥελϦϯά 解決策: 学習前に似た地域を １つのグループとして扱う ΞϝϦΧ ೔ຊ ࣅͨ ૬ޓ࡞༻ ಉ͡άϧʔϓ
35. 35 ஍ҬͷάϧʔϓΫϥελϦϯά 𝒇 𝑥$%& |𝝌𝒕,𝒊,: = 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏& 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 FLUXCUBE: 𝑔*, … 𝑔5$ =K-means UMAP 𝒂𝒊, 𝒃𝒊, 𝑪𝒊 , 𝑖 = [1, … , 𝐿] 各グループ を推定 各地域の反応項のパラメータ 地域の数 ൓Ԡ߲
36. 36 ஍ҬͷάϧʔϓΫϥελϦϯά 𝒇 𝑥$%& |𝝌𝒕,𝒊,: = 𝑎& 𝝌𝒕,𝒊,𝒋 1 − ∑&+ 𝑐&&+ 𝑥$%&+ 𝑏& 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 FLUXCUBE: 𝑔*, … 𝑔5$ =K-means UMAP 𝒂𝒊, 𝒃𝒊, 𝑪𝒊 , 𝑖 = [1, … , 𝐿] Reaction term taylor swift lady gaga beyonce katy perry maroon 5 stevie wonder Group A: Russia, Mexico, Spain, Vietnam, etc. Group B: United Kingdom, Germany, France, etc. Group C: United States, Canada, Australia, etc. Group D: Iran, Pakistan, Brazil, Argentina, etc. Group E: China, Japan, Vietnam, South Africa, etc. 最適なグループ数 (𝑑!)を Model Description Cost (MDL) を⽤いて探索 各グループ を推定 各地域の反応項のパラメータ 地域の数
37. 37 ࣮ݧઃఆ l 13, 26, 52 िؒઌͷ༧ଌ஋ͷਫ਼౓ΛධՁ l ධՁࢦඪ l RMSE, MAE: খ͍͞஋Ͱ͋Δ΄Ͳߴ͍ਫ਼౓ σʔληοτ l Google Trend͔ΒऔಘՄೳͳ10೥ؒ ͷ1िؒ͝ͱͷݕࡧ਺ (7छྨ) ࣮ݧ
38. 38 ࣮ݧ݁Ռ ࣮ݧ
39. 39 ࣮ݧ݁Ռ ࣮ݧ
40. 40 ࣮ݧ݁Ռ ࣮ݧ
41. 41 Ablation Study ࣮ݧ 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& قઅ߲Λ࡟আ ֦ࢄ߲Λ࡟আ ֦ࢄ߲Λఆ਺ʹஔ͖׵͑ ఏҊϞσϧ
42. 42 Ablation Study ࣮ݧ 𝜕𝑥$%& 𝜕𝑡 = 𝒇 𝑥$%& |𝝌𝒕,𝒊,: + ℊ 𝑥$%& |𝝌𝒕 , 𝒕 𝑥$)*%& = 𝐹 𝜕𝑥$%& 𝜕𝑡 + 𝑥$%& ℊ 𝑥)*+|𝝌𝒕 , 𝒕 = + 𝑅𝑁𝑁(1: 𝑡) ⊙ 𝝌𝒕 𝜶 ⊙ 𝝌𝒕 قઅ߲Λ࡟আ ֦ࢄ߲Λ࡟আ ֦ࢄ߲Λఆ਺ʹஔ͖׵͑ ఏҊϞσϧ
43. 43 Case study1: Vod Case Study AppleTV / ESPN / HBO / Hulu / Netflix Sling / Vudu / Youtube Keyword List
44. 44 Case study1: Vod Case Study ૬ޓ࡞༻
45. 45 Case study1: Vod Case Study 影響の流れ
46. 46 Case study2: SNS Case Study Line/facebook/slack/snapchat/ twitter/viber/whatsapp Keyword List
47. 47 Case study2: SNS Case Study ૬ޓ࡞༻
48. 48 Case study2: SNS Case Study 影響の流れ
49. 49 l FLUXCUBE: ൓Ԡ֦ࢄํఔࣜͱੜଶܥΛදݱ͢Δ਺ཧϞσϧ -PULB 7PMUFSB NPEFM ʹجͮ͘ϞσϦϯάख๏ɽೖྗ͞Εͨσʔλʹજࡏత ʹଘࡏ͢Δقઅੑ΍૬ޓ࡞༻ͳͲͷྗֶతͳؔ܎Λநग़ l (PPHMF
    5SFOEΛ༻͍ͨσʔλΛ༻͍ͯɼఏҊϞσϧͷ༗ޮੑΛఏࣔ l ؍ଌσʔλͷഎޙʹଘࡏ͢Δજࡏతͳ૬ޓ࡞༻΍ӨڹͷྲྀΕΛɼਓؒ ͕ղऍՄೳͳܗΛఏڙ ࠓޙͷల๬ l ࣗಈతʹؔ܎ͷ͋ΔΩʔϫʔυͱؔ܎ͷͳ͍ΩʔϫʔυΛ۠ผ͢Δ͜ ͱΛݕ౼ ·ͱΊ