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

word2vecで女性向けQ＆Aサイトを解析してみた

Avatar for tatsushim tatsushim
June 17, 2015
Research
6.4k
0

 word2vecで女性向けQ＆Aサイトを解析してみた

2015/06/10
IVS CTO NightのLTで発表したプレゼン資料です。
word2vecにmamariQ内のテキストを使用してみました。

Avatar for tatsushim

tatsushim

June 17, 2015

More Decks by tatsushim

See All by tatsushim
DockerとAmazon SageMakerで実現した機械学習システムのプロダクション移行
Avatar for tatsushim tatsushim
0
27k
コミュニティサービスにおける機械学習のためのアノテーション
Avatar for tatsushim tatsushim
0
2.2k
日本のママをコンテナで支える
Avatar for tatsushim tatsushim
1
3.1k
コネヒトが考える技術選択の仕方について
Avatar for tatsushim tatsushim
0
22k
コネヒトが考えるサービスづくりに必要な技術とその考え方について
Avatar for tatsushim tatsushim
2
3.2k
独身男性のためのデータドリブン講座
Avatar for tatsushim tatsushim
0
6.9k

Other Decks in Research

See All in Research
ForestCast: Forecasting Deforestation Risk at Scale with Deep Learning
Avatar for SatAI.challenge satai
3
600
A History of Approximate Nearest Neighbor Search from an Applications Perspective
Avatar for Yusuke Matsui matsui_528
1
230
学習型データ構造：機械学習を内包する新しいデータ構造の設計と解析
Avatar for Yusuke Matsui matsui_528
6
4.4k
空間音響処理における物理法則に基づく機械学習
Avatar for NII S. Koyama's Lab skoyamalab
0
250
存立危機事態の再検討
Avatar for Ken Jimbo jimboken
0
260
From Data Meshes to Data Spaces
Avatar for Posedio posedio
PRO
0
540
それ、チームの改善になってますか？ー「チームとは？」から始めた組織の実験ー
Avatar for hirakawa hirakawa51
0
1k
データサイエンティストの業務変化
Avatar for The Japan DataScientist Society datascientistsociety
PRO
0
330
"主観で終わらせない"定性データ活用 ― プロダクトディスカバリーを加速させるインサイトマネジメント / Utilizing qualitative data that "doesn't end with subjectivity" - Insight management that accelerates product discovery
Avatar for 株式会社カミナシ kaminashi
16
23k
「車1割削減、渋滞半減、公共交通2倍」を 熊本から岡山へ＠RACDA設立30周年記念都市交通フォーラム2026
Avatar for Traffic Brain trafficbrain
1
850
2026 東京科学大 情報通信系 研究室紹介 (すずかけ台)
Avatar for Tokyo Tech ICT Dept. icttitech
0
1.3k
Can We Teach Logical Reasoning to LLMs? – An Approach Using Synthetic Corpora (AAAI 2026 bridge keynote)
Avatar for もりし morishtr
1
180

Featured

See All Featured
<Decoding/> the Language of Devs - We Love SEO 2024
Avatar for Nikki Halliwell nikkihalliwell
1
170
The Curious Case for Waylosing
Avatar for Cassini Nazir cassininazir
0
280
The Illustrated Children's Guide to Kubernetes
Avatar for Chris Short chrisshort
51
52k
The Web Performance Landscape in 2024 [PerfNow 2024]
Avatar for Tammy Everts tammyeverts
12
1.1k
How STYLIGHT went responsive
Avatar for nonsquared nonsquared
100
6k
How GitHub (no longer) Works
Avatar for Zach Holman holman
316
150k
Rails Girls Zürich Keynote
Avatar for Gregor Martynus gr2m
96
14k
ラッコキーワード サービス紹介資料
Avatar for ラッコ株式会社 rakko
1
2.8M
Winning Ecommerce Organic Search in an AI Era - #searchnstuff2025
Avatar for Aleyda Solis aleyda
1
1.9k
Building Flexible Design Systems
Avatar for Yesenia Perez-Cruz yeseniaperezcruz
330
40k
The SEO Collaboration Effect
Avatar for Kristina Bergwall kristinabergwall1
0
410
AI Search:
Where Are We & What Can We Do About It?
Avatar for Aleyda Solis aleyda
0
7.2k

Transcript

  1. word2vecͰঁੑ޲͚QˍA αΠτΛղੳͯ͠Έͨ Tatsuro Shimada <[email protected]> tatsushim @ Connehito, Inc

  2. Connehito Inc. ౡాୡ࿕ʢ͠·ͩͨͭΖ͏ʣ • Connehito, Inc CTO • Πϯϑϥ͔Βϑϩϯτ·Ͱ •

    ͋ͱ͓՛ࢠͷങ͍ग़͠ PROFILE @tatsushim 2

  3. ϚϚϦͱ͸ʁ Connehito Inc. 3

  4. Connehito Inc. ϚϚϦjp (❨web)❩ ϚϚϦ2 (❨ΞϓϦ)❩ ϚϚϦKQ / ϚϚϦ2 ϝσΟΞ

    ίϛϡχςΟ 4

  5. Connehito Inc. ࣭໰΁ͷճ౴ ˋ௒ ճ౴͕ͭ͘·Ͱ ෼ҎԼ ΞϓϦͷ଺ࡏ࣌ؒ ෼Ҏ্ ѹ౗తͳαʔϏεͷ੒௕ 

        +VO +VM "VH 4FQ 0DU /PW %FD +BO 'FC લ݄ൺˋ੒௕ ྦྷܭ౤ߘ਺ ೛৷ɾग़࢈Ͱ೰Ήਓͷ ਓʹਓ͕݄̍ΞΫηε 5

  6. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 6 λʔήοτ ࢲ

  7. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 7 ঁੑ λʔήοτ ࢲ

  8. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 8 ঁੑ ϚϚ λʔήοτ ࢲ

  9. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 9 ঁੑ ϚϚ λʔήοτ ࢲ

    ೛৷ ग़࢈

  10. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 10 ঁੑ ϚϚ λʔήοτ ࢲ

    உੑ ೛৷ ग़࢈

  11. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 11 ঁੑ ϚϚ λʔήοτ ࢲ

    உੑ ೛৷ ग़࢈ ಠΓ਎

  12. Connehito Inc. ϚϚϦKQ / ϚϚϦ2 12 ঁੑ ϚϚ λʔήοτ ࢲ

    உੑ ಠΓ਎ ೛৷ ग़࢈ Ͱ͖ͳ͍

  13. ʘ(^o^)ʗ

  14. ʘ(^o^)ʗ Ϣʔβʔͷ͜ͱཧղ͍ͨ͠ʂ

  15. 15

  16. 16 word2vec

  17. None

  18. Connehito Inc. • Tomas Mikolovࢯ
 (౰࣌Google, ݱFacebook)͕ఏҊ • ୯ޠಉ࢜ͷؔ܎ੑΛϕΫτϧͱͯ͠දݱ •

    ୯ޠͷྨࣅ౓ͷܭࢉ΍଍͠Ҿ͖͕Մೳ word2vecͱ͸ʁ 18

  19. 19

  20. 20 ۩ମྫΛݟͯΈΑ͏

  21. word2vecͷ୅දతͳ2ͭͷ࢖͍ํ Connehito Inc. 21

  22. Display similar words Connehito Inc. 22

  23. None

  24. ྨٛޠ

  25. Interesting properties of the word vectors Connehito Inc. 25

  26. Connehito Inc. 26 word2vecͷྫ vector(‘France')

  27. Connehito Inc. 27 word2vecͷྫ - vector(‘Paris') vector(‘France')

  28. Connehito Inc. 28 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) vector(‘France')

  29. Connehito Inc. 29 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France')

  30. Connehito Inc. 30 word2vecͷྫ vector(‘Paris')

  31. Connehito Inc. 31 word2vecͷྫ - vector(‘France') vector(‘Paris')

  32. Connehito Inc. 32 word2vecͷྫ - vector(‘France') + vector(‘Italy’) vector(‘Paris')

  33. Connehito Inc. 33 word2vecͷྫ - vector(‘France') + vector(‘Italy’) = vector('Rome')

    vector(‘Paris')

  34. Connehito Inc. 34 word2vecͷྫ vector(‘king')

  35. Connehito Inc. 35 word2vecͷྫ - vector(‘man') vector(‘king')

  36. Connehito Inc. 36 word2vecͷྫ - vector(‘man') + vector(‘woman’) vector(‘king')

  37. Connehito Inc. 37 word2vecͷྫ - vector(‘man') + vector(‘woman’) = vector('queen')

    vector(‘king')

  38. ੔ཧͯ͠ΈΔ Connehito Inc. 38

  39. Connehito Inc. 39 word2vecͷྫ - vector(‘Paris') vector(‘France')

  40. Connehito Inc. 40 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) vector(‘France')

  41. Connehito Inc. 41 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France')

  42. Connehito Inc. 42 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France')}

  43. Connehito Inc. 43 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ }

  44. Connehito Inc. 44 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ } +

  45. Connehito Inc. 45 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ } } +

  46. Connehito Inc. 46 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ } } ౦ژ +

  47. Connehito Inc. 47 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ } } ౦ژ + =

  48. Connehito Inc. 48 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ } } ౦ژ + = }

  49. Connehito Inc. 49 word2vecͷྫ - vector(‘Paris') + vector(‘Tokyo’) = vector('Japan')

    vector(‘France') ͋Δ౎ࢢΛट౎ͱ͢Δࠃ } } ౦ژ + = }౦ژΛट౎ͱ͢Δࠃ

  50. 50

  51. 51 mamariQͰword2vecͯ͠ΈΔ

  52. None

  53. ୯ޠͷҙຯΛද͢୯ޠ Connehito Inc. 53

  54. Connehito Inc. 54 ʮ଻ಈʯin mamariQ Word: ଻ಈ Word Cosine distance

    ------------------------------------------------------------------------ ೘ҙ 0.527825 ϙίϙί 0.516658 ҧ࿨ײ 0.432082 ಈ͖ 0.430563 ͠Όͬ͘Γ 0.406297 ͙͍ͬͨ͘͢ 0.386457 ಈ͍ 0.383030 ى෬ 0.381906 ϙίο 0.377959

  55. Connehito Inc. 55 ʮυΫϯυΫϯʯin mamariQ Word: υΫϯυΫϯ Word Cosine distance

    ------------------------------------------------------------------------ ຺ଧͭ 0.454460 ϙίο 0.425674 Ͳ͘Ͳ͘ 0.425287 ϐΫο 0.418931 ಥͬுΔ 0.417948 ϐΫϐΫ 0.415464 લଆ 0.413287 ͭͬͺΔ 0.412516

  56. Connehito Inc. 56 ʮ཭ೕ৯ʯin mamariQ Word: ཭ೕ৯ Word Cosine distance

    ------------------------------------------------------------------------ ख͔ͮΈ 0.472250 ͔ͭΈ 0.445568 ॏ౬ 0.432616 ͓͔Ώ 0.425068 τΠτϨ 0.415463 ͨΜͺ࣭͘ 0.412253 ϕϏʔμϊϯ 0.393488 λϯύΫ࣭ 0.392157 ৯ࡐ 0.390477

  57. ୯ޠͱ୯ޠͷؔ܎ੑ Connehito Inc. 57

  58. Connehito Inc. 58 vector(‘ެཱ') ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  59. Connehito Inc. 59 - vector(‘ଉࢠ') vector(‘ެཱ') ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  60. Connehito Inc. 60 - vector(‘ଉࢠ') + vector(‘່’) vector(‘ެཱ') ୯ޠͷ໘ന͍ؔ܎ੑ in

    mamariQ

  61. Connehito Inc. 61 - vector(‘ଉࢠ') + vector(‘່’) = vector('ࢲཱ') vector(‘ެཱ')

    ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  62. 62

  63. 63 ঁͷࢠͷํ͕ࢲཱΛݕ౼͢Δʁ

  64. Connehito Inc. 64 vector(‘νϡʔϋΠ') ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  65. Connehito Inc. 65 - vector(‘͓฼͞Μ') vector(‘νϡʔϋΠ') ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  66. Connehito Inc. 66 - vector(‘͓฼͞Μ') + vector(‘͓෕͞Μ’) vector(‘νϡʔϋΠ') ୯ޠͷ໘ന͍ؔ܎ੑ in

    mamariQ

  67. Connehito Inc. 67 - vector(‘͓฼͞Μ') + vector(‘͓෕͞Μ’) = vector('Ϗʔϧ') vector(‘νϡʔϋΠ')

    ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  68. 68

  69. 69 உੑͷํ͕Ϗʔϧ޷͖ʁ

  70. Connehito Inc. 70 vector(‘ՄѪ͍') ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  71. Connehito Inc. 71 - vector(‘ଉࢠ') vector(‘ՄѪ͍') ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  72. Connehito Inc. 72 - vector(‘ଉࢠ') + vector(‘່’) vector(‘ՄѪ͍') ୯ޠͷ໘ന͍ؔ܎ੑ in

    mamariQ

  73. Connehito Inc. 73 - vector(‘ଉࢠ') + vector(‘່’) = vector('ՄѪ͍') vector(‘ՄѪ͍')

    ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ

  74. Connehito Inc. 74 - vector(‘ଉࢠ') + vector(‘່’) = vector('ՄѪ͍') vector(‘ՄѪ͍')

    ୯ޠͷ໘ന͍ؔ܎ੑ in mamariQ ʊਓਓਓਓਓʊ ʼɹՄѪ͍ɹʻ ʉY^Y^Y^Y
  75. None

  76. ͨͩͷ਌όΧ

  77. ೔ຊޠʹword2vecదԠ͢ΔࡍͷTips Connehito Inc. 77

  78. Connehito Inc. 78 ೔ຊޠʹword2vecదԠ͢ΔࡍͷTips

  79. Connehito Inc. • ࣙॻʹmecab-ipadic-neologdΛ࢖͏
 - Web্ͷݴޠࢿݯ͔Βಘͨ৽ޠʹରԠ 79 ೔ຊޠʹword2vecదԠ͢ΔࡍͷTips

  80. Connehito Inc. • ࣙॻʹmecab-ipadic-neologdΛ࢖͏
 - Web্ͷݴޠࢿݯ͔Βಘͨ৽ޠʹରԠ 80 ೔ຊޠʹword2vecదԠ͢ΔࡍͷTips • ۀքݻ༗ͷϫʔυ͸ͪΌΜͱొ࿥͢Δ


    - Ex. 24w3d = ೛৷͔Β24िͱ3೔໨

  81. Connehito Inc. • ࣙॻʹmecab-ipadic-neologdΛ࢖͏
 - Web্ͷݴޠࢿݯ͔Βಘͨ৽ޠʹରԠ 81 ೔ຊޠʹword2vecదԠ͢ΔࡍͷTips • ύϥϝʔλௐ੔


    - Ex. αʔϏεʹ߹Θͤͨwindow sizeΛ • ۀքݻ༗ͷϫʔυ͸ͪΌΜͱొ࿥͢Δ
 - Ex. 24w3d = ೛৷͔Β24िͱ3೔໨

  82. 82

  83. 83 ݁ہԿʹ࢖͑Δͷ͔ʁ

  84. Connehito Inc. 84 Ԡ༻ઌ

  85. Connehito Inc. • ྨٛޠݕग़ 85 Ԡ༻ઌ

  86. Connehito Inc. • ྨٛޠݕग़ 86 • QˍAίϛχϡχςΟͰ͸Qʹରͯ͠࠷΋ྑ ͍ճ౴ΛܾΊΔͱ͖ͷfeatureʹ࢖͏ͱ͔(࣮ ࡍԠ༻ͨ͠࿦จ͕͋Δ) Ԡ༻ઌ

  87. Connehito Inc. • ྨٛޠݕग़ 87 • QˍAίϛχϡχςΟͰ͸Qʹରͯ͠࠷΋ྑ ͍ճ౴ΛܾΊΔͱ͖ͷfeatureʹ࢖͏ͱ͔(࣮ ࡍԠ༻ͨ͠࿦จ͕͋Δ) •

    ͍͔ͭ͘Ծઆ͕͋ΔͷͰɺ্ख͘ߦͬͨΒ ࿦จʹͯ͠ൃ৴͍͖ͯ͠·͢ Ԡ༻ઌ

  88. ͜͜·Ͱ͕word2vecͷ͓࿩ Connehito Inc. 88

  89. Connehito Inc. • Ҏ্͕ɺword2vecΛ࢖ͬͯΈ͓ͨ࿩Ͱ͢ 89 • IVS CTO NightॳࢀՃͰϫΫϫΫͯ͠·͢ •

    ͜͏͍͏RˍDΛΈΜͳ͍ͭ΍ͬͯɺͲ͏ औΓೖΕͯΔͷ͔஌Γ͍ͨ • ͜ͷޙྑ͔ͬͨΒ͓࿩͍ͤͯͩ͘͞͞ վΊ͓ͯٓ͘͠ئ͍͠·͢ʂ

  90. Connehito Inc. • Ҏ্͕ɺword2vecΛ࢖ͬͯΈ͓ͨ࿩Ͱ͢ 90 • IVS CTO NightॳࢀՃͰϫΫϫΫͯ͠·͢ •

    ͜͏͍͏RˍDΛΈΜͳ͍ͭ΍ͬͯɺͲ͏ औΓೖΕͯΔͷ͔஌Γ͍ͨ • ͜ͷޙྑ͔ͬͨΒ͓࿩͍ͤͯͩ͘͞͞ վΊ͓ͯٓ͘͠ئ͍͠·͢ʂ ͝ਗ਼ௌ༗೉͏͍͟͝·ͨ͠ʂ