PDFファイル 1H3NFC02b 近未来チャレンジセッション「NFC (サバイバル) 異種協調型災害情報支援システム実現に向けた基盤技術の構築 」

The 28th Annual Conference of the Japanese Society for Artificial Intelligence, 2014

1H3-NFC-02b-5

Twitter

ར༻ऀͷ౤ߘ׆ಈʹجͮ͘ఆྔԽख๏ͷ

ࡂ֐৘ใࢧԉγεςϜʹ޲͚ͯͷల๬

Current status and prospects of Twitter users’ quantification method based on their posting

activity for constructing disaster information support system

∗1

দຊ ৻ฏ

∗2

઒ޱ େو

∗3

ௗւ ෆೋ෉

∗1

޿ౡ޻ۀେֶ৘ใֶ෦

Faculty of Applied Information Science, Hiroshima Institute of Technology

∗2

_{޿ౡ޻ۀେֶେֶӃ޻ֶܥݚڀՊ}

Graduate School of Science and Technology, Hiroshima Institute of Technology

∗3

౦ژେֶେֶӃ޻ֶܥݚڀՊ

At the time of the Great East Japan Earthquake, many Tweets of the disaster had posted and Twitter had been effectively-utilized as an infrastructure for sharing disaster information and confirming safety. From now the authors have been addressed the researches to utilize Twitter for disaster under the importance of user classification. Concretely by focusing on Twitter user’s tweeting, replying, and retweeting activities which is assumed to be the source of Twitter’s real time feature, and by numerically-expressing each Twitter user’s activities with a quantification method based on entropy, the Twitter users’ tendency under the disaster and the possibility for user filtering have been examined. This paper shows the summary of our research results previously reported, and expresses the prospects of the quantification method for constructing disaster information support system.

1.

͸͡Ίʹ

౦೔ຊେ਒ࡂൃੜ࣌ɼTwitter͸وॏͳ৘ใަ׵खஈͱͯ͠

ੵۃతʹ׆༻͞Εॏཁͳ໾ׂΛՌͨ͠ɼ·ͨར༻ऀͷ৘ใఏ

ڙ΍ऩूʹߩݙͨ͠[ௗւ14]ɽࡂ֐౰೔ɼଟ͘ͷਓʑ͕҆൱

֬ೝ΍৘ใऩूͳͲͷ໨తͰ༷ʑͳ௨৴खஈͷར༻ΛࢼΈ͍ͯ ͨɽ௨৴ճઢͷඃ֐΍ఀిɼτϥϑΟοΫ૿΍௨৴ن੍ͷͨΊ

ʹి࿩΍ϝʔϧ͕ܨ͕Γʹ͍͘ঢ়گͱͳͬͨҰํͰɼWebӾ

ཡͷύέοτ௨৴͸ൺֱతར༻Ͱ͖Δঢ়گͰ͋ͬͨͨΊɼଟ͘ ͷਓ͕TwitterΛར༻͠ࡂ֐ʹؔ͢Δେྔͳ৘ใΛଈ࣌ऩू ͢Δ͜ͱ͕Ͱ͖ͨɽ·ͨɼճઢʹେ͖ͳෛ୲Λ͔͚Δ͜ͱͳ͘ ৘ใ֦ࢄ͕ՄೳͰ͋ͬͨɽଟ͘͸Ո଒΍஌ਓͷ҆൱ͷ֬ೝɼඃ ࡂ஍ҬͷҩࢣʹΑΔҩྍ૬ஊɼߦ੓ػؔʹΑΔ৘ใൃ৴ͳͲͰ ͋ΓɼϚεϝσΟΞʹཔΒͳ͍৘ใ֦ࢄΛ໨తͱͯ͠ɼඃࡂ஍

಺֎ͷ஍ํ࣏ࣗମ΍ݸਓʹΑͬͯ׆༻͞Ε͍ͯͨ[ࠤʑ໦11]ɽ

ஶऀΒ͸͜Ε·Ͱɼࡂ֐࣌ʹ͓͚ΔTwitterͷ༗ޮͳ׆༻

๏ͷߏஙΛ໨ࢦͯ͠ݚڀΛਐΊ͍ͯΔɽࡂ֐࣌ʹ͓͚Δར༻ऀ

෼ྨͷॏཁੑΛഎܠʹɼTwitterͷϦΞϧλΠϜੑͷݯઘͰ͋

Δͱߟ͑ΒΕΔར༻ऀͷ౤ߘɾฦ৴ɾҾ༻ͷ׆ಈʹண໨ͯ͠ɼ ࡂ֐࣌ͷTwitterͷ࢖ΘΕํͷ෼ੳ΍ར༻ऀͷଐੑʹԠͨ͡ಛ ௃ͷௐࠪΛਐΊ͍ͯΔɽ۩ମతʹ͸ɼ·ͣɼ৘ใྔͷߟ͑ํʹ

ج͍ͮͯաڈͷ౤ߘ׆ಈΛఆྔԽ͢Δ͜ͱʹΑΓ[Ghosh 11]

֤ར༻ऀΛଟมྔఆྔԽ͠ɼTwitterར༻ऀͷ׆ಈΛఆྔతʹ

ղऍ͢ΔͨΊͷख๏ͷߏஙΛߦͬͨɽ࣍ʹɼ౦೔ຊେ਒ࡂલޙ ʹTwitter্ʹ࣮ࡍʹྲྀ௨ͨ͠౤ߘΛର৅ʹ࣮ݧΛߦ͍ɼར ༻ऀଐੑʹԠͨ͡ಛ௃Λ෼ੳ͢Δͱڞʹɼར༻ऀͷࣗಈ൑ผʹ ޲͚ͯͷՄೳੑΛݕূͨ͠ɽຊߘͰ͸ɼ͜Ε·ͰஶऀΒ͕ਐΊ ͖ͯͨऔ૊ͷ֓ཁΛड़΂Δͱڞʹɼࡂ֐৘ใγεςϜߏஙʹ޲ ͚ͯͷࠓޙͷల๬Λड़΂Δɽ

࿈བྷઌ:দຊ৻ฏɼ޿ౡ޻ۀେֶ৘ใֶ෦஌త৘ใγεςϜֶ

Պɼ˟731-5193޿ౡࢢࠤഢ۠ࡾ୐2-1-1, E-Mail: [email protected]

2.

ιʔγϟϧϝσΟΞͷࡂ֐࣌׆༻ʹ޲͚ͯ

ͷपลಈ޲

౦೔ຊେ਒ࡂ࣌TwitterͰ͸ඃࡂ஍ͷঢ়گΛ஌Δਓؒͷ౤

ߘͳͲϚεϝσΟΞ͕ใ͡ͳ͍وॏͳ৘ใ͕਺ଟ͘ྲྀ௨͍ͯ͠ ͕ͨɼࡂ֐ʹؔ͢Δ৘ใΛޮ཰తʹ֫ಘ͢Δٕज़͕ඞཁͰ͋

Δͱߟ͑ΒΕΔɽେن໛ࡂ֐࣌ʹ͓͚ΔSNSʹΑΔۓٸ௨ใ

ͷՄೳੑʹؔ͢Δݕ౼ձͰ͸ɼࡂ֐࣌ʹ͓͍ͯԻ੠௨ใ్͕

ઈ͑ͨ৔߹ʹSNSͳͲͷ৘ใΛ׆༻ͨ͠ۓٸ௨ใͷՄೳੑʹ

͍ͭͯใࠂ͠ɼٕज़త͋Δ͍͸ӡ༻্ͷ՝୊ɼSNSͷ׆༻ํ

๏ͷࡏΓํΛݕ౼͠·ͱΊ͍ͯΔ[૯຿লফ๷ி13]ɽTwitter

্ʹ͸༷ʑͳछྨͷ๲େͳྔͷ౤ߘ͕ৗʹྲྀ௨͍ͯ͠ΔͨΊɼ

ࡂ֐࣌ʹ͓͍ͯTwitterΛԁ׈ʹར༻Ͱ͖ΔΑ͏ʹ͢ΔͨΊ

ʹ͸ɼ໨తʹԠͯ͡౤ߘΛత֬ʹࣗಈ൑ผ͢Δ࢓૊Έ͕ඞཁͰ ͋Δͱड़΂͍ͯΔɽ·ͨɼൃ৴ऀͷҐஔͱࡂ֐ͷద߹ੑɼ৘ใ ͷ৴པੑ޲্Λ՝୊ͱ͍ͯ͋͛ͯ͠Δɽಉ༷ʹɼௗւΒͷऔΓ

૊ΈͰ͸[ௗւ14]ɼ৘ใॲཧʹؔ͢Δ՝୊ͱͯ͠ɼ৘ใͷ࣌

ؒతɾۭؒతͳ֬౓ྼԽ΁ͷରԠɼ৘ใͷ৴པੑอূɼ৘ใͷ ෆ଍ɾܽམ΁ͷରԠɼେن໛ॲཧΛ͍͋͛ͯΔɽ͜ͷதͰ৴པ ੑΛ֬อ͢ΔͨΊͷٕज़՝୊ͱͯ͠ɼϊΠζআڈɼ৘ใ౷߹ɼ σϚͷൃݟͱ๷ࢭɼγϛϡϨʔγϣϯʹΑΔ֬౓ͷݕূ͕ࣔ͞ Ε͍ͯΔ͕ɼࡂ֐࣌ʹ͓͚ΔιʔγϟϧϝσΟΞΛର৅ͱͨ͠ ଟ͘ͷݚڀ͸ɼҎ্Ͱࣔ͞ΕͨϑϨʔϜϫʔΫͷ΋ͱͰ਱ߦ͞ Ε͍ͯΔɽ

3.

ஶऀΒͷऔΓ૊ΈͷҐஔ෇͚

ࡂ֐࣌ʹ͓͍ͯॏཁͱͳΔ৘ใ͸ɼࡂ֐ʹڧؔ͘܎͢Δར ༻ऀ΍ඃࡂऀࣗ਎ʹΑͬͯൃ৴͞ΕΔ৔߹͕ଟ͍ɽ͜ΕΒར༻ ऀͷૌ͑Λత֬ʹநग़͢ΔͨΊʹ͸ɼਓ͕ൃ৴ͨ͠৘ใͷࣝผ ΍ར༻ऀͷଐੑʹԠͨ͡બผɼ͋Δ͍͸ࣗಈ౤ߘϓϩάϥϜ

(bot)ʹΑΓൃ৴͞ΕΔ౤ߘͷݕ஌͕ॏཁͳ՝୊Ͱ͋Δͱߟ͑

ΒΕΔɽ·ͨɼޡ৘ใͷ֦ࢄ͸ɼීஈͷTwitterʹ͸ݟΒΕ

The 28th Annual Conference of the Japanese Society for Artificial Intelligence, 2014

ͳ͍ಛผͳ࢖͍ํΛͨ͠ར༻ऀ΍ɼଞͷϦιʔεΛࣗಈҾ༻

͢Δbot͕ݪҼͷͻͱͭͰ͋ͬͨͱߟ͑ΒΕ͍ͯΔɽҎ্Α

Γɼbotͷݕग़ͷΈͳΒͣར༻ऀͷ׆ಈܗଶͷ೺Ѳ͸ɼޡ৘ใ

΁ରԠ͢ΔͨΊͷॏཁͳ՝୊Ͱ͋Δͱߟ͑ΒΕΔɽࡂ֐࣌ͷ

Twitterར༻ऀͷಛ௃ΛܭࢉՄೳͳܗͰ༰қʹ೺ѲͰ͖Ε͹ɼ

botͷݕ஌ɾআ֎ɼ·ͨɼࣄ࣮Λ஌͍ͬͯΔඃࡂऀͷ੠ͷऔಘ

ʹ׆༻ՄೳͰ͋Δͱߟ͑ΒΕΔɽͦͯͦ͠ͷ݁Ռͱͯ͠ɼޡ৘ ใͷగਖ਼ɾޡ৘ใ֦ࢄͷ཈੍ʹ޲͚ͯͷ׆༻΍ɼඃࡂͨ͠஍Ҭ ͷਓୡͷੜͷ੠ͷऔಘ͕ظ଴Ͱ͖Δɽར༻ऀͷબผɼͱΓΘ͚

botݕग़ʹண໨͢Ε͹ɼͦͷٕज़ͱͯ͠ɼ౤ߘ಺༰ʹج͍ͮͨ

ػցతϑΟϧλϦϯά͕ߟ͑ΒΕΔɽ͔͠͠ͳ͕Βɼ౤ߘ಺༰ ʹج͍ͮͨ৘ใબผ͸ैདྷ੩తͳ৘ใΛର৅ʹ͓ͯ͠Γɼ͞Β ʹɼ͜ΕΒ͸ݱࡏٕज़తʹ੒ख़ͷҬʹ͋ΔɽιʔγϟϧϝσΟ Ξͷ࣮࣌ؒੑʹద߹ͨ͠৘ใબผख๏Λߏங͢ΔͨΊʹ͸ɼ࣭ తख๏Ҏ֎͔Βͷ؍఺ʹج͍ͮͨܭࢉखॱ͕ඞཁͰ͋Δͱߟ͑ ΒΕ͍ͯΔɽҎ্എܠͷ΋ͱͰɼஶऀΒ͸ར༻ऀͷ౤ߘ׆ಈʹ ج͍ͮͨఆྔԽख๏ʹண؟͠ɼଟมྔʹΑͬͯදݱ͞Εͨࡂ֐ ࣌ʹ͓͚Δར༻ऀͷఆྔղऍͱͦͷ෼ੳɼར༻ऀ෼ྨʹ޲͚ͯ

ͷՄೳੑݕূʹऔΓ૊ΜͰ͖ͨ[઒ޱ13a]ɽ

4.

͜Ε·Ͱͷ੒Ռ

Twitterར༻ऀͷఆྔԽख๏͸ɼॲཧ೔Λى఺ͱͯ͠લ਺೔

Λղੳର৅ظؒͱ͠ɼظؒ಺ͷར༻ऀͷ௨ৗ౤ߘɼRetweetɼ

Replyͷ׆ಈύλʔϯ͔ΒΤϯτϩϐʔΛࢉग़͢Δɽ·ͣɼຊ

ݚڀͷख๏ͷجૅͱͳͬͨGhoshΒͷख๏ͱͷൺֱΛߦͬͨɽ

GhoshΒͷख๏Ͱ͸ɼपғͷRetweet׆ಈΛ෼ੳର৅ͱͯ͠ ͍ΔɽҰํɼຊݚڀͰ͸ࡂ֐࣌ͷ׆༻͕໨తͰ͋Γɼࣄྫʹର ͢Δଈ࣌ੑΛอͭ͜ͱͷॏཁੑΛ౿·͑ɼΤϯτϩϐʔͷࢉग़ ʹ͔͔ΔܭࢉྔΛߟྀͯ͠ར༻ऀ͕౤ߘͨ͠શ౤ߘͷΈΛର ৅ͱͨ͠ɽ࣮ݧͷ݁Ռɼର৅σʔλͷ࣭͕େ͖͘ҟͳΔʹ΋ؔ ΘΒͣɼ྆ख๏ͱ΋਺஋ͷࠩҧ͸͋Δ΋ͷͷར༻ऀͷಛ௃෇͚ ͕֓ͶՄೳͰ͋Δ͜ͱ͕֬ೝ͞Εͨɽ·ͨɼ਒ࡂ࣌ʹ͓͚Δࠃ ಺ͷTwitter৘ใʹ͓͍ͯ΋GhoshΒ͕ࣔͨ͠ಛ௃ͱಉ༷ͷ ܏޲͕͋ΓɼఆྔԽʹΑΔ෼ྨ͕ՄೳͰ͋Δ͜ͱ͔ΒΤϯτϩ ϐʔʹΑΔఆྔԽख๏͸ීวੑΛ࣋ͭ͜ͱ͕ࣔࠦ͞Εͨɽ

࣍ʹɼղੳର৅ظؒͷ௕͕͞Τϯτϩϐʔ࣌ܥྻʹ༩͑ΔӨ ڹΛ෼ੳͨ͠ɽͦͷ݁Ռɼղੳର৅ظؒΛ࠷΋୹͘ઃఆͨ͠৔ ߹ʹಥൃతࣄ৅ͷӨڹ͕࠷΋ݦஶʹදΕ͍ͯΔ͜ͱɼҰํͰɼ ෼ੳظ͕ؒ޿͍΄ͲΤϯτϩϐʔͷ஋ͷมಈͷ൓Ԡ΍ϘϥςΟ ϦςΟ͕খ͘͞ͳΔ͜ͱΛ໌Β͔ʹͨ͠ɽಥൃతࣄ৅ͷݕग़ͱ ར༻ऀଐੑͷ෼ྨ͸૬൓ͷؔ܎ʹ͋Δ͜ͱΛ౿·͑ͯɼಥൃత ࣄ৅ͷݕग़ͱར༻ऀଐੑͷ෼ྨΛಉ࣌ʹߦ͏ͨΊͷख๏ΛఏҊ ͨ͠ɽ͜͜Ͱ͸ଞʹɼΤϯτϩϐʔͷִ࣌ؒؒΛ༷ʑͳ؍఺͔ Β༩͑Δ͜ͱʹΑΓɼैདྷ๏ͷಛ௃Λࣦ͏͜ͱͳ࣍͘ݩΛ֦ு ͢Δ͜ͱʹ੒ޭͨ͠ɽ

Ҏ্ͷऔΓ૊ΈͷதͰ͸ఏҊख๏ͷࣗಈ൑ผ΁ͷԠ༻Մೳ ੑΛݕূ͓ͯ͠Γɼ࣮ݧͷ݁Ռ͔Β͸ɼଟมྔఆྔԽʹΑΓར ༻ऀଐੑʹԠͨࠩ͡ҧΛදݱͰ͖ͨ͹͔ΓͰͳ͘ɼࣗಈ൑ผͷ Ԡ༻ՄೳੑΛࣔ͢͜ͱ͕Ͱ͖ͨɽ

5.

·ͱΊͱࠓޙͷల๬

ຊߘͰ͸ɼ͜Ε·ͰஶऀΒ͕ਐΊ͖ͯͨऔ૊ͷ֓ཁΛड़΂ ͨɽࡂ֐৘ใγεςϜߏஙʹ޲͚ͯͷࠓޙͷల๬ͱͯ͠ɼଟม ྔͰදݱ͞Εͨར༻ऀͷ஋͔Βར༻ऀಉ࢜ͷڑ཭Λࢉग़͠ɼ͜

ͷ஋Λ׆༻ͨ͠WebγεςϜͷ։ൃͱԠ༻Λߟ͍͑ͯΔɽ۩

ମతʹ͸ɼ౤ߘ׆ಈͱ౤ߘཤྺͷؔ܎ੑΛಉ࣌ʹ೺Ѳ͢Δͨ ΊɼΤϯτϩϐʔͷ؍఺͔Β౤ߘ׆ಈͷ͍ۙෳ਺ར༻ऀͷ౤ߘ

ਤ1: λάΫϥ΢υγεςϜͷҰྫ

಺༰ΛλάΫϥ΢υͰදݱ͢ΔγεςϜͷߏஙΛݕ౼͍ͯ͠Δ

(ਤ1ࢀর)ɽ·ͣɼଟ༷ͳ౤ߘ಺༰ͷར༻ऀΛ൑ผ͢ΔͨΊɼ

༧Ί༻ҙ͞Εͨࡂ֐Ωʔϫʔυͷॏෳ཰ΛJaccard܎਺Λ༻

͍ͯར༻ऀ͝ͱʹಛ௃෇͚Δɽࣅͨ౤ߘ׆ಈͷར༻ऀಉ࢜͸͋ Δఔ౓ͷڞىੑΛ͕࣋ͭ໌֬ʹ౤ߘ಺༰͕ॏෳ͸͍ͯ͠ͳ͍఺ Λ౿·͑ɼपғͷڑ཭ͷ୹͍ར༻ऀ͔Βऩूͨ͠ςΩετσʔ λΛ༻͍ͯλάΫϥ΢υΛੜ੒͢Δɽλά͸पลͷར༻ऀͷ౤ ߘཤྺ͔Β඼ࢺ৘ใ໊͕ࢺͰ͋Δ΋ͷΛର৅ͱ͢ΔɽҎ্ͷৄ ࡉ͸౰೔ൃදͰࣔ͢ɽ

ࢀߟจݙ

[Ghosh 11] Ghosh et al.: Entropy-based Classification of ‘Retweeting’ Activity on Twitter, Proc. of KDD workshop on Social Network Analysis (SNA-KDD), http://arxiv.org/pdf/1106.0346.pdf (2011).

[઒ޱ13a] ઒ޱ ଞ: Twitterར༻ऀͷఆྔԽख๏ͷࡂ֐࣌׆

༻ʹ޲͚ͯͷ༗ޮੑධՁ,ୈ15ճIEEE޿ౡࢧ෦ֶੜγ

ϯϙδ΢Ϝ࿦จू, pp.474-477 (2013).

[ࠤʑ໦11] ࠤʑ໦: ֦େΛଓ͚ΔTwitterͷ਒ࡂʹ͓͚Δ׆༂ ͱࠓޙͷ՝୊ɼAD STUDIES, Vol.36, pp.20-24 (2011).

[૯຿লফ๷ி13] ૯຿লফ๷ி: େن໛ࡂ֐࣌ʹ͓͚Δ

ι ʔ γϟϧɾωοτ ϫ ʔ Ω ϯ άɾα ʔ Ϗ ε ʹ Α Δ ۓ ٸ ௨ ใ ͷ ׆ ༻ Մ ೳ ੑ ʹ ؔ ͢ Δ ݕ ౼ ձ ใ ࠂ ॻ (2013)ɼ

http://www.fdma.go.jp/neuter/topics/houdou/h25/ 2503/250327 1houdou/02 houkokusho.pdf,

2013/12/10ࢀর

[ௗւ14] ௗւ ଞɼҟछڠௐܕࡂ֐৘ใࢧԉγεςϜ࣮ݱʹ޲

͚ͨج൫ٕज़ͷߏஙɼਓ޻஌ೳֶձ࿦จࢽɼVol.29ɼNo.1ɼ

pp.113-119 (2014)ɽ