Graph Neural Networks のビジネス応用可能性  heterogeneous graph と論文再現性について

(SBQI/FVSBM/FUXPSLTͷϏδωεԠ༻Մೳੑ  IFUFSPHFOFPVTHSBQIͱ࿦จ࠶ݱੑʹ͍ͭͯ

಺ڮ ݎࢤ uchi_k @__uchi_k__ About me yuni, inc. ୅ද nlpaper.challenge
ӡӦ Freelance Machine Learning ɹɹɹɹɹEngineer / Researcher former ະ౿16, ژେ৘ใӃ, FreakOut Machine Learning Engineer

About yuni େاۀ͔ΒελʔτΞοϓɺݚڀػؔͳͲ͔Βػցֶशؔ࿈ͷडୗ։ൃ Λߦ͖ͬͯ·ͨ͠ σʔλυϦϒϯͳ΋ͷͮ͘Γࣄۀͱͯ͠ɺΦϯϥΠϯ໰਍ʹΑΔύʔι φϥΠζ৸۩ͷ੡࡞΋΍͍ͬͯͨΓ͠·͢ ࠓ೔͸ɺIFUFSPHFOFPVTHSBQIͷ࿩Λ͠Α͏ͱࢥ͍ͬͯ·͢ ࡢ೥຤૑ۀͨ͠ɺγʔυظʹ͋ΔελʔτΞοϓʢࣾһ໊ʣͰ͢ ػցֶशºϚʔέςΟϯάྖҬͰαʔϏε։ൃΛ͍ͯ͠·͢

ࠓ೔͓࿩͢Δ͜ͱ ϚʔέςΟϯάºػցֶश ࣮ੈքͷάϥϑͷੑ࣭ )FUFSPHFOFPVTHSBQI͔Βͷػցֶशͱ࣮Ԡ༻ ࿦จͷ࠶ݱੑʹ͍ͭͯ ·ͱΊͱ(//ք۾ʹؔͯ͠ࢥ͏͜ͱ

.BSLFUJOHʷ.BDIJOF-FBSOJOH ϚʔέςΟϯάྖҬʹ͸਺ଟ͘ͷϨΠϠʔ͕͋Γɺ ػցֶश͕ར༻͞Ε͍ͯΔ΋ͷ΋ଟ͋͘Δɻ https://www.smartinsights.com/managing-digital-marketing/marketing-innovation/15-applications-artiﬁcial-intelligence-marketing/ άϥϑσʔλ͕ଟ͘ɺকདྷతʹ(//͕༗ޮͳͷͰ͸ʁ ͬ͘͟ΓͰ͕͢ɺҰൠతͳϑϨʔϜϫʔΫʹԊͬͯλεΫΛ঺հ͠·͢

3"$& • 3FBDIΠϯό΢ϯυ ◦ 4/4ࢪࡦͳͲͷίϯςϯπϚʔέςΟϯάɺ4&0ɺϝσΟΞͳͲͰ ϢʔβʔΛࣗࣾʹ༠ಋ͢Δ ◦ ػցֶश͕࢖ΘΕ͍ͯΔͷ͸ҎԼͷΑ͏ͳλεΫ ◦ ίϯςϯπϚʔέςΟϯάͷޮՌଌఆ
◦ ίϯςϯπ΍*OUFOUͷΩϡϨʔγϣϯʢϦίϝϯυʣ ◦ ӡ༻ܕ޿ࠂ • "DUؔ৺Λ࣋ͨͤΔ ◦ ༠ಋͨ͠જࡏސ٬ʹରͯ͠ద੾ͳࢪࡦɾಋઢΛઃܭ͢Δ ◦ ػցֶश͕࢖ΘΕ͍ͯΔͷ͸ɺ ◦ ϦʔυείΞϦϯάʢજࡏސ٬ͷॱҐ෇͚ʣ ◦ ސ٬ߦಈɾίϯςϯπ࠷దԽʢ-10ͳͲʣ ◦ λʔήςΟϯά޿ࠂ

3"$& • $POWFSUߪങʹಈ͔͢ ◦ ڵຯΛ࣋ͬͯ͘Εͨސ٬ʹߪೖ·ͰͷಋઢΛҾ͘ ◦ ػցֶश͕࢖ΘΕ͍ͯΔͷ͸ɺ ◦ ಈతՁ֨ઃఆʢηʔϧͳͲͷࢪࡦ΋ؚΉʣ ◦
ಋઢઃܭʢνϟοτϘοτͳͲʣ ◦ ϦλʔήςΟϯά޿ࠂ • &OHBHFސ٬ͱΑ͍ؔ܎Λங͘ ◦ ܧଓతʹߪೖ΍ొ࿥Λͯ͠΋Β͏ͨΊɺྑ࣭ͳؔ܎Λҡ࣋͢Δ ◦ ػցֶश͕࢖ΘΕ͍ͯΔͷ͸ɺ ◦ ΤϯήʔδΦʔτϝʔγϣϯʢ୭ʹɺԿΛɺ͍ͭɺͷࣗಈԽʣ ◦ ΧελϚʔαϙʔτʢϘοτʣ

6TFS(FOFSBUFE$POUFOUT 6($ऩू 4/4֦ࢄ ݕࡧ ޿ࠂࢪࡦ ՄࢹԽ 4/4Λى఺ʹͨ͠ίϯςϯπϚʔέͷޮՌଌఆΛࢧԉ

ϝλύεΛ௨ͯۙ͠઀ؔ܎͕ఆٛͰ͖Δ NFUBQBUIWFD4DBMBCMF3FQSFTFOUBUJPO-FBSOJOHGPS)FUFSPHFOFPVT/FUXPSLT ϚʔέςΟϯάʹ͓͚ΔHSBQI ࣮ੈքͷάϥϑ͸IFUFSPHFOFPVTͳ΋ͷ͕ଟ͍

)FUFSPHFOFPVT(SBQI"UUFOUJPO/FUXPSL 9JBP8BOH #FJKJOH6OJWFSTJUZPG1PTUTBOE5FMFDPNNVOJDBUJPOT FUBM888 DJUBUJPOT ݱ࣮ੈքͷIFUFSPHFOFPVTHSBQIʹର͠ɺϝλύεΛྡ઀ؔ܎ͱͨ͠ ֊૚BUUFOUJPOʹΑΔख๏Ͱ4P5"Λୡ੒ )FUFSPHFOFPVTHSBQIʹ͓͚ΔϝλύεΛհͨ͠ྡ઀ؔ܎Λఆٛ

)FUFSPHFOFPVT(SBQI"UUFOUJPO/FUXPSL 9JBP8BOH #FJKJOH6OJWFSTJUZPG1PTUTBOE5FMFDPNNVOJDBUJPOT FUBM888 DJUBUJPOT ϝλύεۙ๣ϊʔυϨϕϧͰͷBUUFOUJPOͱɺ ϝλύεϨϕϧͰͷBUUFOUJPOͷ֊૚తͳBUUFOUJPOΛఆٛ OPEFMFWFMBHHSFHBUJPOͰϝλύεۙ๣Λॏཁ౓ʹԠͯ͡ू໿͢Δ ڞ௨ۭؒ΁ͷࣹӨI@J Τοδͷॏཁ౓FЇ@JK
ਖ਼نԽͨ͠Τοδॏཁ౓ЋЇ@JK ϝλύεʹجͮ͘ಛ௃ຒΊࠐΈ[Ї@J

)FUFSPHFOFPVT(SBQI"UUFOUJPO/FUXPSL 9JBP8BOH #FJKJOH6OJWFSTJUZPG1PTUTBOE5FMFDPNNVOJDBUJPOT FUBM888 DJUBUJPOT λεΫʹΑͬͯϝλύεͷॏཁ౓͸ҟͳΔͨΊɺ ϝλύεͷॏཁ౓ΛTFNBOUJDBUUFOUJPOͱͯ͠ಋೖ TFNBOUJDMFWFMBHHSFHBUJPOͰϝλύεΛॏཁ౓ʹԠͯ͡ू໿͢Δ ୯૚.-1ͰOPEFFNCFEEJOHΛ ඇઢܗม׵͠ɺຒΊࠐΈͷྨࣅ౓ͱ͠
ͯॏཁ౓Λܭࢉ ਖ਼نԽ ϝλύεॏཁ౓ͰBHHSFHBUJPO

)FUFSPHFOFPVT(SBQI"UUFOUJPO/FUXPSL 9JBP8BOH #FJKJOH6OJWFSTJUZPG1PTUTBOE5FMFDPNNVOJDBUJPOT FUBM888 DJUBUJPOT ϊʔυϨϕϧɺϝλύεϨϕϧͷॏཁ౓͕ݟΕΔͷͰɺͦͷ͋ͨΓͷղ ऍੑ͕ٻΊΒΕΔ৔߹ʹ࢖͑ͦ͏ ϝλύεͷઃܭ͕ඞཁͳͷͰɺ͋Δఔ౓ͷυϝΠϯ஌ࣝ͸ඞཁͦ͏ ࣮ݧ݁ՌΛݟΔݶΓϝλύεؒͷॏཁ౓͕͍͍ͩͨಉ͡ʹͳͬͯ͠· ͍ɺ݁ՌTFNBOUJDGVTJPO͕͋·Γҙຯͳ͍Α͏ʹݟ͍͑ͯΔͷͰɺ
΋ͬͱϝλύε͕όϦΤʔγϣϯ๛͔Ͱॏཁ౓͕େ͖͘ҟͳΔλεΫͰ ࢼͯ͠Έ͍ͨ

)FUFSPHFOFPVTHSBQIͷ࣮Ԡ༻ • )FUFSPHFOFPVT(SBQI/FVSBM/FUXPSLTGPS.BMJDJPVT "DDPVOU%FUFDUJPO ◦ ;JRJ-JV "OU'JOBODJBM4FSWJDFT(SPVQ FUBM $*,.
DJUBUJPOT ◦ "MJQBZͰѱҙͷ͋ΔΞΧ΢ϯτΛݕग़͢ΔͨΊʹɺσόΠε΍ΞΫ ςΟϏςΟ͔ΒͳΔIFUFSPHFOFPVTHSBQI͔ΒͷػցֶशΛߦ͏ • .FUBQBUIHVJEFE)FUFSPHFOFPVT(SBQI/FVSBM/FUXPSLGPS *OUFOU3FDPNNFOEBUJPO ◦ 4IBPIVB'BO #FJKJOH6OJWFSTJUZPG1PTUTBOE 5FMFDPNNVOJDBUJPOT FUBM,%% DJUBUJPOT ◦ Ϣʔβʔ͕5BPCBPΛ։͍ͨͱ͖ʹɺաڈͷߦಈ͔ΒࣗಈతʹΫΤϦ ʢJOUFOUʣΛϦίϝϯυ͢Δ

%//ਪનք۾Ͱى͍ͬͯ͜Δ໰୊ ৽نख๏ͷग़ݱස౓͕ߴ͍σʔληοτ͕ଟذʹΘͨΔ͜ͱ ͔Βɺ4P5"͕ඇৗʹ௥͍ʹ͘͘ͳ͍ͬͯΔ 3FD4ZT#FTU1BQFSͰɺۙ೥ͷ%//ܥਪનख๏Λ ࠶ݱ࣮૷͠ൺֱͨ݁͠Ռɺݹయతͳख๏ʹ͢Βউͯͳ͍΋ͷ͕΄ͱΜͲͩͬͨ ,%% 888 3FD4ZT 4*(*3ͷ%//ؔ࿈࿦จຊͷ௥ࢼͰɺ ࠶ݱͰ͖ͨͷ͕൒෼ҎԼͷຊ͔͠ͳ͔ͬͨ
࠶ݱͰ͖ͯ΋ɺ΄ͱΜͲ͕ݹయతͳख๏ʹෛ͚ͯ͠·ͬͨ ਖ਼͍͠ϕʔεϥΠϯʢݹయత͕ͩڧྗͳख๏ʣΛਖ਼͘͠ௐ੔͠ɺ ൺֱ͠Α͏ɻ࣮૷ɺલॲཧɺϋΠύϥΛެ։͠Α͏ɻ

(//ͷʮਖ਼͍͠ϕʔεϥΠϯʯΛ࡞ΔऔΓ૊Έ • "'BJS$PNQBSJTPOPG(SBQI/FVSBM/FUXPSLTGPS(SBQI $MBTTJpDBUJPO 'FEFSJDP&SSJDB FUBM *$-3 DJUBUJPOT ◦ %($//
%J⒎1PPM &$$ (*/ (SBQI4"(&ͷͭͷϞσϧΛɺͭ ͷϕϯνϚʔΫͰ࠶࣮ݧ ◦ (//Λߏ଄ʹͱΒΘΕͳ͍ϕʔεϥΠϯͱൺֱ͢Δ͜ͱͰɺߏ଄৘ ใ͕׆༻͞Ε͍ͯΔ͔Λݕূ • 0QFO(SBQI#FODINBSL%BUBTFUTGPS.BDIJOF-FBSOJOHPO (SBQIT 8FJIVB)V FUBMBS9JW DJUBUJPOT ◦ େن໛ʢ࠷େԯϊʔυɺԯΤοδʣͰଟ༷ͳλεΫɾྖҬΛΧ όʔͨ͠ϕϯνϚʔΫσʔληοτΛఏҊ ◦ 044తʹӡ༻͞Ε͍ͯΔ

(//ͷʮਖ਼͍͠ϕʔεϥΠϯʯΛ࡞ΔऔΓ૊Έ • #FODINBSLJOH(SBQI/FVSBM/FUXPSLT 7JKBZ1SBLBTI %XJWFEJ FUBMBS9JW DJUBUJPOT ◦ طଘख๏͸খن໛ͳσʔλͰධՁ͞Ε͍ͯΔ͜ͱ͕ଟ͘ɺҰൠԽՄೳ ͳΞʔΩςΫνϟ͔൑ఆͮ͠Β͍
◦ HSBQISFHSFTTJPODMBTTJpDBUJPO OPEFDMBTTJpDBUJPO MJOL QSFEJDUJPOͷͭͷλεΫʹ͍ͭͯɺϕϯνϚʔΫͱͯ͠࢖༻͢Δʹ ;͞Θ͍͠σʔληοτΛબఆ ◦ ($/T 8-(//Tͷطଘख๏Ͱ࠶ݱ࣮ݧΛͨ͠

#FODINBSLJOH(SBQI/FVSBM/FUXPSLT BS9JW %XJWFEJ 7JKBZ1SBLBTIBOE+PTIJ $IBJUBOZB,BOE-BVSFOU 5IPNBTBOE#FOHJP :PTIVBBOE#SFTTPO 9BWJFS DJUBUJPOT தن໛σʔλͰͷάϥϑճؼʗ෼ྨɺϊʔυϦϯΫ༧ଌͳͲͷλεΫͰɺ
طଘख๏ͷ࠶ݱ࣮૷Λൺֱͨ͠ɻ طଘख๏Ͱ࢖ΘΕ͍ͯΔσʔληοτ͕খن໛͗͢Δ໰୊ "'BJS$PNQBSJTPOPG(SBQI/FVSBM/FUXPSLTGPS(SBQI$MBTTJpDBUJPO 'FEFSJDP&SSJDB FUBM *$-3

طଘख๏ͷ࠶ݱ࣮૷Λൺֱͨ͠ɻ (//T 8-(//Tͷ࣮ݧϑϩʔΛఆٛ

طଘख๏ͷ࠶ݱ࣮૷Λൺֱͨ͠ɻ άϥϑߏ଄Λແࢹͯ͠.-1ͰϞσϧԽ͢ΔͱείΞ͕௿͍ͷͰɺάϥ ϑߏ଄ͷϞσϦϯά͕ඞਢͷλεΫͰ͸͋Γͦ͏ 8-(//T͸ɺ8-UFTUͱಉ౳ͷࣝผೳྗΛ࣋ͭ΋ͷͷɺ࠷ۙͷ 8-(//΍3JOH(//͸($/ΛείΞͰԼճΔ

(//࿦จͷ࠶ݱੑ σʔληοτ͕খن໛ͩͬͨΓҰൠԽͮ͠Β͔ͬͨΓͰɺ4P5" ͕௥͍ʹ͍͘ͱ͍͏఺͸%//ਪનͱಉ͕ͩ͡ɺ͜Ε͸͋Δఔ ౓͠ΐ͏͕ͳ͍໘΋͋Δ άϥϑΧʔωϧΛ࢖ͬͨख๏ͳͲɺݹయతख๏ͱͷͪΌΜͱ͠ ͨൺֱ΋ݟͯΈ͍ͨɾɾɾ ίʔυ͕ެ։͞Ε͍ͯΔ࿦จ͕ଟ͘ɺடং͕อͨΕ͍ͯΔΑ͏ ʹײ͡·͢ʢࢲݟͰ͢ʣ

ύωϧσΟεΧογϣϯ IFUFSPHFOFPVTHSBQIͷख๏ʹ͸ଞʹͲΜͳ΋ͷ͕͋Δ ͷʁ ͦ΋ͦ΋άϥϑߏ଄ͷຒΊࠐΈʹ͸ͲΜͳछྨ͕͋Δʁ ࠷ۙҰ൪໘ന͍ͱࢥͬͨ࿦จɾऔΓ૊Έ͸ͳΜͰ͔͢ʁʢෳ਺ ͋ͬͯ΋େৎ෉Ͱ͢ʂʣ

Graph Neural Networks のビジネス応用可能性  heterogeneo...

Graph Neural Networks のビジネス応用可能性  heterogeneous graph と論文再現性について

uchi_k

More Decks by uchi_k

Other Decks in Programming

Featured

Transcript

(SBQI/FVSBM/FUXPSLTͷϏδωεԠ༻Մೳੑ  IFUFSPHFOFPVTHSBQIͱ࿦จ࠶ݱੑʹ͍ͭͯ

಺ڮ ݎࢤ uchi_k @__uchi_k__ About me yuni, inc. ୅ද nlpaper.challenge

ࠓ೔͓࿩͢Δ͜ͱ ϚʔέςΟϯάºػցֶश ࣮ੈքͷάϥϑͷੑ࣭ )FUFSPHFOFPVTHSBQI͔Βͷػցֶशͱ࣮Ԡ༻ ࿦จͷ࠶ݱੑʹ͍ͭͯ ·ͱΊͱ(//ք۾ʹؔͯ͠ࢥ͏͜ͱ

3"$& • 3FBDIΠϯό΢ϯυ ◦ 4/4ࢪࡦͳͲͷίϯςϯπϚʔέςΟϯάɺ4&0ɺϝσΟΞͳͲͰ ϢʔβʔΛࣗࣾʹ༠ಋ͢Δ ◦ ػցֶश͕࢖ΘΕ͍ͯΔͷ͸ҎԼͷΑ͏ͳλεΫ ◦ ίϯςϯπϚʔέςΟϯάͷޮՌଌఆ

3"$& • $POWFSUߪങʹಈ͔͢ ◦ ڵຯΛ࣋ͬͯ͘Εͨސ٬ʹߪೖ·ͰͷಋઢΛҾ͘ ◦ ػցֶश͕࢖ΘΕ͍ͯΔͷ͸ɺ ◦ ಈతՁ֨ઃఆʢηʔϧͳͲͷࢪࡦ΋ؚΉʣ ◦

6TFS(FOFSBUFE$POUFOUT 6($ऩू 4/4֦ࢄ ݕࡧ ޿ࠂࢪࡦ ՄࢹԽ 4/4Λى఺ʹͨ͠ίϯςϯπϚʔέͷޮՌଌఆΛࢧԉ

ϝλύεΛ௨ͯۙ͠઀ؔ܎͕ఆٛͰ͖Δ NFUBQBUIWFD4DBMBCMF3FQSFTFOUBUJPO-FBSOJOHGPS)FUFSPHFOFPVT/FUXPSLT ϚʔέςΟϯάʹ͓͚ΔHSBQI ࣮ੈքͷάϥϑ͸IFUFSPHFOFPVTͳ΋ͷ͕ଟ͍

)FUFSPHFOFPVT(SBQI"UUFOUJPO/FUXPSL 9JBP8BOH #FJKJOH6OJWFSTJUZPG1PTUTBOE5FMFDPNNVOJDBUJPOT FUBM888 DJUBUJPOT ݱ࣮ੈքͷIFUFSPHFOFPVTHSBQIʹର͠ɺϝλύεΛྡ઀ؔ܎ͱͨ͠ ֊૚BUUFOUJPOʹΑΔख๏Ͱ4P5"Λୡ੒ )FUFSPHFOFPVTHSBQIʹ͓͚ΔϝλύεΛհͨ͠ྡ઀ؔ܎Λఆٛ

)FUFSPHFOFPVTHSBQIͷ࣮Ԡ༻ • )FUFSPHFOFPVT(SBQI/FVSBM/FUXPSLTGPS.BMJDJPVT "DDPVOU%FUFDUJPO ◦ ;JRJ-JV "OU'JOBODJBM4FSWJDFT(SPVQ FUBM $*,.

(//ͷʮਖ਼͍͠ϕʔεϥΠϯʯΛ࡞ΔऔΓ૊Έ • "'BJS$PNQBSJTPOPG(SBQI/FVSBM/FUXPSLTGPS(SBQI $MBTTJpDBUJPO 'FEFSJDP&SSJDB FUBM *$-3 DJUBUJPOT ◦ %($//

(//ͷʮਖ਼͍͠ϕʔεϥΠϯʯΛ࡞ΔऔΓ૊Έ • #FODINBSLJOH(SBQI/FVSBM/FUXPSLT 7JKBZ1SBLBTI %XJWFEJ FUBMBS9JW DJUBUJPOT ◦ طଘख๏͸খن໛ͳσʔλͰධՁ͞Ε͍ͯΔ͜ͱ͕ଟ͘ɺҰൠԽՄೳ ͳΞʔΩςΫνϟ͔൑ఆͮ͠Β͍

#FODINBSLJOH(SBQI/FVSBM/FUXPSLT BS9JW %XJWFEJ 7JKBZ1SBLBTIBOE+PTIJ $IBJUBOZB,BOE-BVSFOU 5IPNBTBOE#FOHJP :PTIVBBOE#SFTTPO 9BWJFS DJUBUJPOT தن໛σʔλͰͷάϥϑճؼʗ෼ྨɺϊʔυϦϯΫ༧ଌͳͲͷλεΫͰɺ

#FODINBSLJOH(SBQI/FVSBM/FUXPSLT BS9JW %XJWFEJ 7JKBZ1SBLBTIBOE+PTIJ $IBJUBOZB,BOE-BVSFOU 5IPNBTBOE#FOHJP :PTIVBBOE#SFTTPO 9BWJFS DJUBUJPOT தن໛σʔλͰͷάϥϑճؼʗ෼ྨɺϊʔυϦϯΫ༧ଌͳͲͷλεΫͰɺ

#FODINBSLJOH(SBQI/FVSBM/FUXPSLT BS9JW %XJWFEJ 7JKBZ1SBLBTIBOE+PTIJ $IBJUBOZB,BOE-BVSFOU 5IPNBTBOE#FOHJP :PTIVBBOE#SFTTPO 9BWJFS DJUBUJPOT தن໛σʔλͰͷάϥϑճؼʗ෼ྨɺϊʔυϦϯΫ༧ଌͳͲͷλεΫͰɺ

ύωϧσΟεΧογϣϯ IFUFSPHFOFPVTHSBQIͷख๏ʹ͸ଞʹͲΜͳ΋ͷ͕͋Δ ͷʁ ͦ΋ͦ΋άϥϑߏ଄ͷຒΊࠐΈʹ͸ͲΜͳछྨ͕͋Δʁ ࠷ۙҰ൪໘ന͍ͱࢥͬͨ࿦จɾऔΓ૊Έ͸ͳΜͰ͔͢ʁʢෳ਺ ͋ͬͯ΋େৎ෉Ͱ͢ʂʣ

Graph Neural Networks のビジネス応用可能性 heterogeneo...

Graph Neural Networks のビジネス応用可能性 heterogeneous graph と論文再現性について

More Decks by uchi_k

Other Decks in Programming

Featured

Transcript

Graph Neural Networks のビジネス応用可能性  heterogeneo...

Graph Neural Networks のビジネス応用可能性  heterogeneous graph と論文再現性について