© LINE Data Labs ΢ΣϒاۀͷඇݚڀऀϙδγϣϯͰ ߦ͏αΠΤϯε ਓ޻஌ೳֶձ΢ΣϒαΠΤϯεݚڀձΦʔϓϯηϛφʔ vol. 6 2018/05/26 ߴޱ ଠ࿕ LINE גࣜձࣾ Data Labs

© LINE Data Labs ࣗݾ঺հ ߴޱ ଠ࿕ʢ͔͙ͨͪ ͨΖ͏ʣ LINE Data Labs σʔλαΠΤϯςΟετʢ2017/10 -ʣ ΞΧσϛΞͰͷܦྺ 2010 - 2013 ౦ژେֶେֶӃ ৘ใཧ޻ֶܥݚڀՊ ത࢜՝ఔ 2013 - 2016 ࠃཱ৘ใֶݚڀॴ ಛ೚ݚڀһ 2016 - 2017 ৘ใ௨৴ݚڀػߏ ςχϡΞτϥοΫݚڀһ ΞΧσϛΞͰͷઐ໳෼໺ ωοτϫʔΫՊֶ ର৅ɿ࣮ࣾձͷͭͳ͕Γߏ଄ʢιʔγϟϧάϥϑɺ΢Σϒʣ खஈɿཧ࿦Ϟσϧ + σʔλ෼ੳ Network data from Rocha et al., PLOS Comput. Biol. 7, e1001109 (2011)

© LINE Data Labs ΞΧσϛΞͰͷݚڀ׆ಈ 2 2 2 1 1 1 3 3 4 4 3 4 ● ࣮σʔλ෼ੳ ● ఻೻Ϟσϧʢײછ঱ɺ৘ใ֦ࢄʣͷཧ࿦Ϟσϧղੳ ࣌ؒతʹมԽ͢ΔιʔγϟϧɾωοτϫʔΫͷɿ ࣌ࠁ ● ࠪಡ෇͖࿦จ ܭ22ຊ
 ʢ૯Ҿ༻ճ਺ 300+ ճ in Google Scholarʣ ● ࠃࡍձٞ NetSci 2017 ϓϩάϥϜҕһ ● ࠃࡍڞಉݚڀɿε΢ΣʔσϯɺϕϧΪʔɺϑϥϯεɺΠΪϦε

© LINE Data Labs DATA LABS ͱ͸ Data Labs Data Science Machine Learning Engineer Planner ● LINE ؔ࿈֤αʔϏεͷσʔλΛԣஅతʹѻ͏ઐ໳૊৫ ● 2016/04 ઃஔ LINE ϑΝϛϦʔαʔϏε ࣄۀ෦ Ϗδωε্ͷ՝୊ σʔλΛ׆༻ͨ͠ ղܾࡦ

© LINE Data Labs 1. ؔ܎ݚڀ෼໺ʹ͓͚ΔΞΧσϛΞ
 ͔ΒϏδωε΁ͷҠ੶ͷಈ޲ 2. ͍ͪ “ݚڀऀ” ͔ΒΈͨΞΧσϛΞ ͱ΢Σϒاۀ
 2-1. ҧ͏఺
 2-2. ڞ௨͢Δ఺ 3. σʔλαΠΤϯεݚڀͱ΢Σϒاۀ ͱͷ਌࿨ੑ Agenda

© LINE Data Labs ΞΧσϛΞ͔ΒϏδωε΁ͷҠ੶ͷಈ޲ ● ؔ܎ݚڀ෼໺ɿωοτϫʔΫՊֶ ● ΞϝϦΧͰ͸ 2010 ೥ࠒ͔Βগͳ͔Βͣྫ͕͋Δ Ϗδωε΁Ҡ੶ͨ͠ւ֎ݚڀऀͷࣄྫʢҰ෦ɺॱෆಉʣ ● A ڭत: C େֶ → Y ࣾ → M ࣾ ● B ।ڭत: M େֶ → F ࣾ ● C ।ڭत: S େֶ → P ࣾ (part-time) ● D ݚڀһ: O େֶ → F ࣾ ਪଌͱͯ͠ҰൠʹݴΘΕΔҠ੶ͷཧ༝ ● େن໛ͳ࣮σʔλΛѻ͑Δ͜ͱ ● ଴۰ͷྑ͞

© LINE Data Labs ΞΧσϛΞ͔ΒϏδωε΁ͷҠ੶ͷಈ޲ 㱺 ඞͣ͠΋ͦΕ͚͕ͩཧ༝Ͱ͸ͳ͍ͷͰ͸ʁ ● ؔ܎ݚڀ෼໺ɿωοτϫʔΫՊֶ ● ΞϝϦΧͰ͸ 2010 ೥ࠒ͔Βগͳ͔Βͣྫ͕͋Δ Ϗδωε΁Ҡ੶ͨ͠ւ֎ݚڀऀͷࣄྫʢҰ෦ɺॱෆಉʣ ਪଌͱͯ͠ҰൠʹݴΘΕΔҠ੶ͷཧ༝ ● େن໛ͳ࣮σʔλΛѻ͑Δ͜ͱ ● ଴۰ͷྑ͞ ● A ڭत: C େֶ → Y ࣾ → M ࣾ ● B ।ڭत: M େֶ → F ࣾ ● C ।ڭत: S େֶ → P ࣾ (part-time) ● D ݚڀһ: O େֶ → F ࣾ

© LINE Data Labs ͍ͪ “ݚڀऀ” ͔ΒΈͨΞΧσϛΞ ͱ΢Σϒاۀ

© LINE Data Labs ● ΞΧσϛΞ = ౷ܭ෺ཧΑΓͷωοτϫʔΫՊֶ ● Ϗδωεɹ = ݱॴଐ෦ॺ ● ݴ໌Λ্هҎ্ʹҰൠԽ͢Δҙਤ͸͋Γ·ͤΜ ΞΧσϛΞͱϏδωεͱͷҧ͍

© LINE Data Labs ● ΞΧσϛΞ = ౷ܭ෺ཧΑΓͷωοτϫʔΫՊֶ ● Ϗδωεɹ = ݱॴଐ෦ॺ ● ݴ໌Λ্هҎ্ʹҰൠԽ͢Δҙਤ͸͋Γ·ͤΜ ΞΧσϛΞͱϏδωεͱͷҧ͍ ҧ͏͜ͱͩΒ͚

© LINE Data Labs ● ΞΧσϛΞ = ౷ܭ෺ཧΑΓͷωοτϫʔΫՊֶ ● Ϗδωεɹ = ݱॴଐ෦ॺ ● ݴ໌Λ্هҎ্ʹҰൠԽ͢Δҙਤ͸͋Γ·ͤΜ ΞΧσϛΞͱϏδωεͱͷҧ͍ ΞΧσϛΞ Ϗδωε 1.υΩϡϝϯτɾίʔυͷ
 ެ։ൣғ ڞಉݚڀऀͲ͏͠ ݚڀࣨ಺ جຊతʹશࣾ಺
 ʢػີϨϕϧґଘʣ 2. νʔϜߏ੒ ओʹಉ෼໺ͷݚڀऀ ϓϩδΣΫτ୯ҐͰ ଟछଟ༷ 3. ࡞ۀऀͷΫϨδοτ ඞਢ ڧௐ͞Εͳ͍ 4. ֶज़తͳ৽نੑ ඞਢ ॏࢹ͞Εͳ͍

© LINE Data Labs ΞΧσϛΞͱϏδωεͱͷڞ௨఺ 1. ࿦ཧతࢥߟͱ՝୊ղܾͷํ๏࿦ɾϑϨʔϜϫʔΫͷॏཁੑ ޷ح৺ ઌߦݚڀαʔϕΠ ݚڀσΟεΧογϣϯ طଘɾ৽نͷ ਺ཧతख๏ ౷ܭతख๏ CONTENTS ೔ৗͷ՝୊ҙࣝ ࣄۀଆ͔Βͷґཔ ݪஶ࿦จ ֶձൃද ϓϨθϯςʔγϣϯ վળͷ࣮૷ ● Ϗδωε΁ͷస޲Ͱ͸ઐ໳஌͕ࣝॏࢹ͞Ε͕͕ͪͩɺҰ࿈ͷαΠ ΫϧΛεϜʔζʹ܁Γฦ࣮͠ߦͰ͖Δ͜ͱͦ͜ॏཁ ● ৘ใऩूɾ໰୊ઃఆɾίϛϡχέʔγϣϯͷೳྗ Πϯϓοτ Ξ΢τϓοτ ઐ໳஌ࣝ ΞΧσϛΞ Ϗδωε

© LINE Data Labs ݱঢ়ཧղ ৽نͷاը ΞΧσϛΞͱϏδωεͱͷڞ௨఺ 2. ݁Ռͷ࠶ݱੑ͕ॏཁ ޮՌͷࣄޙ෼ੳ αʔϏε࣮૷ ࣄલ෼ੳ ςετͷઃܭ ݱঢ়ཧղ ৽نͷاը ● ෼ੳ݁Ռ͸௕ظతͳҙࢥܾఆʹӨڹΛٴ΅͢ ● ِཅੑɾِӄੑͷΤϥʔΛ੍ޚ͢ΔͨΊʹ౷ܭͷ஌͕ࣝඞཁ ● ୲౰ऀ͕มΘͬͯ΋ܧଓͰ͖Δ࢓૊ΈԽ΋ඞཁ ౷ܭతख๏ͷཞ༻ɾޡ༻ͷ໰୊΋ڞ௨ ● P-Hacking ● ዞҙతͳՄࢹԽʹىҼ͢ΔϛεϦʔυ ● ೔ৗޠతͳදݱʹΑΔղऍͷᐆດԽ

© LINE Data Labs σʔλαΠΤϯεݚڀͱ
 ΢Σϒاۀͱͷ਌࿨ੑ

© LINE Data Labs ະ஌ͷࣄ৅ʹ͍ͭͯ ୳ٻ͍ͨ͠ ݚڀऀͷࢤ޲ ΢Σϒاۀͷಛੑ αʔϏεɾϢʔβʔɾ ࣾձ͕มԽ͢ΔݶΓ ະ஌ͷࣄ৅͹͔Γ ࠷৽ͷ / ৽نͳ ਺ཧతख๏Λ׆༻ ͍ͨ͠ ม਺ͷ૬ؔͰ͸ͳ͘ ϝΧχζϜ͕஌Γ͍ͨ ఏҊͨ͠಺༰͕ࣾձʹ ༩͑ΔΠϯύΫτΛ ݟ͍ͨ ࣄۀʹߩݙ͢ΔݶΓʹ͓͍ͯ਌࿨ੑ͸ߴ͍ ίϯτϩʔϧ͞Εͨ
 ϥϯμϜԽςετ͕࣮ ࢪՄೳ ఏҊ͕࣮૷͞Εͯ
 Ϣʔβʔʹར༻͞ΕΔ ·Ͱͷϥά͕୹͍ ͦͷख๏ʹΑͬͯ ॴ๬ͷ੒Ռ͕ୡ੒͞Ε ΔͳΒ͹ OK

© LINE Data Labs ཧ૝తͳݚڀࣄྫ Ugander et al., “Structural diversity in social contagion” PNAS 109, 5962 (2012) ● ๭ SNS ͷϢʔβʔσʔλΛ෼ੳ ● ओ݁Ռɿ͋ΔϢʔβʔʹͱͬͯɺ༑ͩͪಉ࢜ͷͭͳ͕Γߏ଄͕
 - ট଴Λड͚ͯ SNS ʹࢀՃ͢Δ֬཰ʢrecruitmentʣ
 - ܧଓͯ͠ SNS Λར༻͢Δ֬཰ʢengagementʣ
 ʹޮ͘ 1. ֶज़తʹՁ஋ͷ͋Δ஌ݟʢιʔγϟϧάϥϑͷܗ੒ϓϩηεʣ 2. SNS ͷϏδωεతʹ΋༗ҙٛ 3. ͋ΓಘΔ༑ͩͪಉ࢜ͷͭͳ͕Γߏ଄͸ແ਺ͷՄೳੑ
 → Ϗοάσʔλ͕͋ͬͯॳΊͯ෼ੳՄೳ ྑ͍ͱ͜Ζ

© LINE Data Labs ૬ؔͰ͸ͳ͘ϝΧχζϜ͕஌Γ͍ͨ ؍࡯ʮιʔγϟϧάϥϑͰͭͳ͕ͬͨ̎ਓ͕ಉ͡ঢ়ଶʹมԽͨ͠ʯ ࣾձత఻೻
 (social contagion) ྨࣅऀબ޷
 (homophily) ڞ௨ͷ֎෦ཁҼ Shalizi & Thomas, “Homophily and contagion are generically cofounded in observational social network studies” Social Methods Res. 40, 211 (2011) ● ؍࡯σʔλͷΈ͔ΒωοτϫʔΫతͳҼՌΛಛఆ͢Δ͜ͱ͸ɺ
 ະ؍ଌɾ؍ଌෆೳͳม਺͕͋ΔݶΓඇৗʹ೉͍͠ ● ݴ͍׵͑Ε͹ɺద੾ʹઃܭɾ౷੍͞ΕͨϥϯμϜԽςετ͕ඞཁ

© LINE Data Labs ֶज़తͳ৽نੑͷॏཁ౓ ੒Ռ͕ධՁ͞ΕΔ ࣌ؒεύϯ ඞਢ ॏࢹ͠ͳ͍ ௕ظ ୹ظ ཧ૝తͳ
 ΞΧσϛΞ ݱ࣮ͷ
 ΞΧσϛΞʁ ݱঢ়ͷ ࣗݾೝࣝ Ϗδωεͷ ෼ੳλεΫ

© LINE Data Labs ·ͱΊ ● େن໛ͳ࣮σʔλΛѻ͑Δ͜ͱ ● ଴۰ͷྑ͞ ͚ͩͰ͸ͳ͘ ● ՝୊ղܾͷϑϨʔϜϫʔΫͱ࠶ݱੑॏࢹͷ࢟੎͕ධՁ͞ΕΔ͜ͱ ● ະ஌ͷࣄ৅ʹ͍ͭͯ୳ٻͰ͖Δ͜ͱ ● ม਺ͷ૬ؔͰ͸ͳ͘ϝΧχζϜΛ஌Δ͜ͱ͕Ͱ͖ΔՄೳੑ ● ఏҊͨ͠಺༰͕ࣾձʹ͋ͨ͑ΔΠϯύΫτΛݟΒΕΔ͜ͱ σʔλαΠΤϯεؔ࿈ͷݚڀऀ͕΢Σϒاۀ΁ҠΔಈػ͚ͮ ● ݸਓ͝ͱͷࢤ޲΍૊৫͝ͱͷ؀ڥʹඇৗʹେ͖ͳ෼ࢄ͕͋Δ ● ʮΞΧσϛΞ vs. ϏδωεʯͷΑ͏ͳա౓ʹ୯७Խͨ͠ೋݩ࿦͸ ͋·Γ༗ӹͰͳ͍ ݸਓతʹେࣄͩͱࢥ͏͜ͱ