リクルートグループの現場事例から見るAI/ディープラーニング技術のビジネス活用と“A3RT”外部公開の裏側

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϦΫϧʔτάϧʔϓͷݱ৔ࣄྫ͔ΒݟΔ "*σΟʔϓϥʔχϯάٕज़ͷϏδωε׆༻ͱ l"35z֎෦ެ։ͷཪଆ גࣜձࣾϦΫϧʔτςΫϊϩδʔζϏοάσʔλ෦ϓϩμΫτ։ൃάϧʔϓ Ԟా༟थ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ࢖ΘΕΔ΋ͷΛ૑Δ αʔϏεΛ࢖͏ਓʹ αʔϏεΛ࡞Δਓʹ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE w ϦΫϧʔτ͸ଟछଟ༷ͳαʔϏεΛఏڙ͍ͯ͠·͢ w ࡶࢽ΍ωοτϝσΟΞͳͲଟ༷ͳαʔϏεΛఏڙ͍ͯ͠·͢ ϦΫϧʔτͱ͍͏ձࣾʹ͍ͭͯ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϦΫϧʔτςΫϊϩδʔζͷཱͪҐஔ U D D BS R H
S S w ϦΫϧʔτςΫϊϩδʔζ͸ɺϦΫϧʔτάϧʔϓͷ*5ɾωοτϚʔέςΟϯάྖҬ ͷςΫϊϩδʔ։ൃΛ୲͏ձࣾ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE & PM T D T T B PM
T R T B R T I TB I TB R T B T I TB PM T B I TB JD TB JD TB B Ϗοάσʔλղੳ෦໳ͷ૊৫ମ੍ʹ͍ͭͯ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE Ϗοάσʔλ෦ʹ͓͚ΔϓϩμΫτ 10 5 1 0 ., 1
.,0 2 1. 22 1 0 . . . oC y vA H L T L e d a s u r th g 5i d R l c p n p a 9 . .,. . ,, 1 e d AC h c igh C e R T L 2 0

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE Ϗοάσʔλ෦ʹ͓͚ΔϓϩμΫτ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϦΫϧʔτͷػցֶशʹΑΔιϦϡʔγϣϯ׆༻ͷ ෑډΛԼ͛ɺࣄۀͷαʔϏεͷՁ஋ΛߴΊΔͨΊͷ ൚༻తͳػցֶश"1*αʔϏε܈

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE w ࣾ಺֤αʔϏεʹରͯ͠ػցֶशιϦϡʔγϣϯΛఏڙ w ϨίϝϯυɾਪનγεςϜɺը૾ೝࣝɺݪߘੜ੒ɺݪߘߍӾɺจষཁ໿ɺԻ੠ೝ ࣝɺର࿩ϘοτɺϢʔβଐੑਪఆɺʜʜ "35 Ξʔτ
ʹؚ·ΕΔ֤ػೳ

"MM3JHIU3FTFSWFE$PQZSJHIU D 3FDSVJU5FDIOPMPHJFT$P -UE ɹɹϨίϝϯυɾਪન"1* աڈͷϢʔβཤྺʹج͍ͮͨϨίϝϯυߏஙͷࣗಈԽ - - w ར༻͢ΔσʔλΛఆܕԽ͠Ϩίϝϯυͷಋೖ࣌ؒ΍ίετΛ୹ॖ
w ͜Ε·ͰݸผαʔϏεʹಛԽͯ͠࡞ΒΕ͍ͯͨϨίϝϯυػೳͷ։ൃΛɺ౷Ұతʹ ѻ͏σʔλεΩʔϜΛࡦఆͯ͠ن֨Խ w ಋೖোนΛݮΒͯ͠ͲͷΑ͏ͳαʔϏεͰ΋ؾܰʹಋೖͰ͖ΔΑ͏ʹαϙʔτ

"MM3JHIU3FTFSWFE$PQZSJHIU D 3FDSVJU5FDIOPMPHJFT$P -UE ɹɹɹɹը૾ղੳ"1* w σΟʔϓϥʔχϯάͰ࠷΋੒ޭͨ͠෼໺ w ೥ͷը૾ೝࣝίϯϖ *-473$
ʹ͓͚Δ ύϥμΠϜγϑτҎޙͷੈք w ࣗಈӡసͳͲͷجૅͱͳΔେ੾ͳٕज़ w ࣮ϓϩμΫτ΁ͷԠ༻ྫ w ෺ମࣝผ w ෆద੾ը૾ݕग़ w λά෇༩ w ը૾ೝࣝʹΑΔϨίϝϯυ 1 / 0 3%# 4*" &- 30% 10% 60% 50, !) 2+ &- . 'Deep Learning ( CNN1)$ ൚༻తͳϩδοΫͰ໨త͝ͱʹಛԽͨ͠ը૾ղੳॲཧ

"MM3JHIU3FTFSWFE$PQZSJHIU D 3FDSVJU5FDIOPMPHJFT$P -UE ɹɹɹݪߘࣗಈੜ੒"1* େྔͷݪߘ৘ใΛֶशͯ͠ɺػցʹΑΔݪߘͷࣗಈ࡞੒ w ֶशͨ͠ݪߘʹ͍ۙจষΛ࡞੒ɾαδΣετ w ͳΔ΂ࣗ͘વͳ೔ຊޠͰɺֶशͨ͠ݪߘʹ͍ۙจষΛੜ੒
w ݪߘ࡞੒ऀͷෛ୲ݮ΍࡞ۀޮ཰Խɺख़࿅ऀͱಉ͡Α͏ͳจষΛ৽ਓ͕ॻ͚ΔΑ͏ͳ αϙʔτΛػց͕୲͏

"MM3JHIU3FTFSWFE$PQZSJHIU D 3FDSVJU5FDIOPMPHJFT$P -UE ɹɹɹɹݪߘߍӾ"1* ϧʔϧϕʔεͱػցֶशͷϋΠϒϦουʹΑΔݪߘߍӾ w ϧʔϧϕʔεʹΑΔνΣοΫ w ೔෇ͱ༵೔ͷ੔߹ੑ
w ॅॴͷؒҧ͍ w /(ϫʔυ w ػցֶशʹΑΔνΣοΫ w ׽ࣈ΍ॿࢺͳͲͷޡࣈ୤ࣈ w ࠩผදݱͱͳΓ͏Δදݱ w ૊Έ߹ΘͤʹΑΔ/(දݱ ʮࠃຽ͸੫ۚΛऩΊΔʯ ʮೲΊΔʯͰ͸ ͳ͍Ͱ͔͢ʁ w ޡࣈ୤ࣈ΍ෆద੾දݱͷࣗಈݕग़ w 8FCܝࡌݪߘͷߍӾʹ͔͔Δ޻਺࣌ؒΛ࡟ݮ͢Δ͜ͱͰɺࠓͷݪߘ࡞੒ମ੍Ͱѻ͏ ͜ͱͷͰ͖Δ૯ྔΛ૿΍͢

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ౷Ұͨ͠"1*ن֨ʹΑΔޮ཰Խ API API API
w ൚༻Խ w Ϟσϧ΍ٕज़Λ൚༻Խ͠ڞ༗ Ͱ͖ΔΑ͏ʹ͢Δ w ͦΕͧΕͷࣄۀσʔλͷಛੑ ʹ߹Θͤͨνϡʔχϯά͸ඞ ཁͱͳΔ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE "351VCMJD"1*ͱͯࣾ͠֎ʹެ։! IUUQTBSUSFDSVJUUFDIDPKQ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE w 1VCMJD"1*ࣾ಺Ͱ஝ੵٕͨ͠ज़ΛແঈͰ֎෦ެ։ w ҰൠϢʔβͷํ͕ར༻Ͱ͖ΔΑ͏ͳٕज़Λز͔ͭ બग़͠ɺ֎෦༻ʹνϡʔχϯά্ͨ͠Ͱఏڙ w ఏڙ͢Δػೳछྨ
w 5FYU$MBTTJpDBUJPO"1*ɿςΩετ෼ྨ w 5FYU4VHHFTU"1*ɿςΩετੜ੒αδΣετ w 1SPPGSFBEJOH"1*ɿݪߘߍӾ w 5BML"1*ɿνϟοτϘοτ w -JTUJOH"1*ɿϨίϝϯυ w *NBHF*OqVFODF"1*ɿը૾ೝࣝ "35͕ఏڙ͢Δػցֶश"1*܈

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE 5FYU$MBTTJpDBUJPO"1*ςΩετ෼ྨ { "status": 0, "message": "ok", "classes": [
{ "probability": 0.958804, "label": "ITΤϯδχΞʢγεςϜ։ൃɾSEɾΠϯϑϥʣ" }, [...] ] } τϐοΫ΁ͷॴଐ֬཰஋ τϐοΫ໊ʢϥϕϧʣ $ curl -X POST https://api.a3rt.recruit-tech.co.jp/text_classification/v1/ classify -d "apikey=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA" -d model_id=default -d "text='γεςϜͷاը͔Β։ൃɾӡ༻·Ͱ෯޿ؔ͘ΘΕ·͢ɻ'"

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE 5FYU4VHHFTU"1*ςΩετੜ੒ɾαδΣετ { "status": 0, "message": "ok", "suggestion": [
"ͷ্ʹݘ͕࠲͍ͬͯ·͢ɻ", "͕̏಄ཱ͍ͬͯ·͢ɻ", "ͷԣͷํ͕ݟ͑·͢ɻ" ] $ curl "https://api.a3rt.recruit-tech.co.jp/text_suggest/v1/predict? apikey=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA&previous_description=അ" ੜ੒จΛ֨ೲͨ͠഑ྻ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE 1SPPGSFBEJOH"1*จষߍӾ { "status": 1, "message": "pointed out", "alerts":
[ { "checkedSentence": "γεςϜ ͷ <<ؾ>> ֨ ͔Β ։ൃ ɾ ӡ༻ ·Ͱ ෯޿͘ ؔΘΕ ·͢ ɻ", "alertCode": 1, "alertDetail": "unnatural", "word": "ؾ", "rankingScore": 0.16431238837473616 }, [...] ] } ࢦఠՕॴΛͰࣔͨ͠จ ࢦఠͷछྨ΍ৄࡉ ࢦఠ୯ޠ ࢦఠͨ͠୯ޠͷٙΘ͠͞Λࣔ͢ࢦඪ $ curl "https://api.a3rt.recruit-tech.co.jp/proofreading/v1/typo? apikey=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA&sentence=γεςϜͷؾ͔֨Β։ൃɾӡ༻·Ͱ ෯޿ؔ͘ΘΕ·͢ɻ"

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE 5BML"1*νϟοτϘοτ { "status": 0, "message": "ok", "results": [
{ "perplexity": 0.12975786883239396, "reply": "੖ΕͯΑ͔ͬͨͰ͢" } ] } $ curl -X POST https://api.a3rt.recruit-tech.co.jp/talk/v1/smalltalk -F "apikey=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA" -F "query=͍͍ఱؾͰ͢Ͷ" ༧ଌੑೳʢύʔϓϨΩγςΟʣ Ԡ౴ςΩετ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE -JTUJOH"1*Ϩίϝϯυ VTFS@JE JUFN@JE UJNF@TUBNQ VTFS@ JUFN@" VTFS@
JUFN@% VTFS@ JUFN@) VTFS@ JUFN@# VTFS@ JUFN@" VTFS@ JUFN@) VTFS@ JUFN@$ *OQVU 0VUQVU VTFS@JE TDPSF@ TDPSF@ ʜ VTFS@ JUFN@$ JUFN@" ʜ VTFS@ JUFN@% JUFN@) ʜ VTFS@ JUFN@& JUFN@1 ʜ VTFS@"ˠ%ˠ) VTFS@#ˠ"ˠ) VTFS@$

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE *NBHF*OqVFODF"1*ը૾ೝࣝ { "status": 0, "message": "ok", "result": {
"score": 3.729642152786255 } } ը૾ͷ࠾఺݁Ռʢd఺ʣ $ curl -X POST -F apikey=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA -F [email protected] -F predict=1 https://api.a3rt.recruit-tech.co.jp/ image_influence/v1/meat_score

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE w ظ଴ͱ໨త w ଟछଟ༷ͳۀछɺଟ͘ͷར༻ऀʹ࢖͍ͬͯͨͩ͘͜ͱʹΑΔϑΟʔυόοΫ w ʮσʔλΛूΊΔʯͰ͸ͳ͘ʮσʔλ͕ू·Δʯ w
ΦʔϓϯΠϊϕʔγϣϯ w ۙ೥ͷػցֶशΛऔΓר͘ϓϩμΫτ͸ෳࡶԽˠ୯ҰͷػೳΛෳ਺૊Έ߹Θͤͯ ෳࡶͳ࢓૊ΈΛ࣮ݱ͢Δ͜ͱ͕ඞཁ w ը૾ೝࣝ͢Δ͚ͩɺจষ෼ྨ͢Δ͚ͩͰ͸ෆे෼ w طଘͷγεςϜͱͲͷΑ͏ʹ༥߹ͤ͞Δ͔ɺػցֶशͰ࣮ݱͰ͖ͳ͍෦෼Λิ ͏͔ w ͦΕͧΕͷػೳΛ͢΂ͯࣗલͰ༻ҙ͢ΔͷͰ͸ͳ͘ɺطʹଘࡏ͢Δ"1*Λ׆༻͢Δ ͜ͱͰɺૉૣ͍։ൃ͕ՄೳʹͳΔ "35ެ։ͷૂ͍ͱɺͦͷઌͷະདྷ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϦΫϧʔτྲྀ"*σΟʔϓϥʔχϯάϏδωε׆༻ͷצॴ w ໰୊Λղܾ͢Δ΋ͷͰͳ͚Ε͹͍͚ͳ͍ w ʮͱΓ͋͑ͣࢼͯ͠Έͨʯ͚ͩͰ͸Ϗδωεͱ͸ݴ͑ͳ͍ w ࣮ࡍʹ࢖ͬͯ΋ΒΘͳ͚Ε͹Ձ஋͕ͳ͍ w
ιϦϡʔγϣϯͱͯ͠ఏڙ͠ɺಋೖ͢Δͱ͜Ζ·Ͱ΍Γൈ͘ w ސ٬ʹίετΛ͔͚Δ͚ͩͷՁ஋͕͋Δͱೲಘͯ͠΋Β͏ w Ϗδωεͱͯ͠εέʔϧ͠ͳ͚Ε͹͍͚ͳ͍ w ݶΒΕͨਓࡐ࣌ؒͰɺΑΓଟ͘ͷར༻ऀΛ֫ಘ͢Δ ͜Ε·ͰͷγεςϜ։ൃͱ͸ҧͬͨߟ͑ํ΍ํ๏Ͱ Ϟϊͮ͘ΓΛ͠ͳ͚Ε͹͍͚ͳ͍

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ྫɿػցֶशͷϓϩμΫτͷढ͍ ػցֶशɿ ٕज़తෛ࠴ͱ͍͏ߴ͍རࢠ͕෇͖·ͱ͏ΫϨδοτΧʔυ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϦΫϧʔτྲྀ"*σΟʔϓϥʔχϯάϏδωε׆༻ͷצॴ w ࿩ͤΔ͜ͱ w ࢲͨͪͷ໨ͷલʹ͋Δ໰୊ΛɺͲ͏΍ͬͨΒػցֶशͷٕज़Λ࢖ͬͯղܾʹಋ͘ ͜ͱ͕Ͱ͖Δ͔ʁ w ߴ͍རࢠΛܭըతʹฦࡁ͍ͯͨ͘͠Ίͷํ๏
w Ϗδωεͷਓ͕ؒػցֶशΛͲ͏ଊ͍͑ͯΔ͔ w ࿩ͤͳ͍͜ͱ w Ͳ͏΍ͬͨΒ"*ਓ޻஌ೳϏδωεͰҰࢁ౰ͯΒΕΔ͔ w Ͳ͜ͷاۀେֶͱڞಉͰ։ൃ͢Ε͹ྑ͍͔

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϦΫϧʔτྲྀ"*σΟʔϓϥʔχϯάϏδωε׆༻ͷצॴ w ݕ౼ฤ ໰୊Λਖ਼͘͠ݟۃΊΔ ظ଴஋Λௐ੔͢Δ
௕ظతͳ൐૸Λ֮ޛ͢Δ w ࣮૷ฤ ػցֶश෦෼ΛγεςϜຊମ͔Β෼཭ͤ͞Δ γϯϓϧͳํ๏͔Βࢼ͍ͯ͘͠ ݁ՌΛਖ਼͘͠ଌΔ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ໰୊Λਖ਼͘͠ݟۃΊΔ w ػցֶशʹΑͬͯԿΛ͠ͳ͚Ε͹͍͚ͳ͍ͷ͔ʁΛ໰͍ଓ͚Δ w ࠷ऴతͳ໨తʹࢸΔͨΊʹɺͲͷΑ͏ͳ໰୊Λղ͘ͷ͔ʁΛߟ͑Δ w ݱ৔Ͱ͸໰୊͸ৗʹมΘΓଓ͚Δʢࢢ৔΍χʔζͷมԽʣ w
໰୊Λղܾ͢ΔͨΊʹ৽ͨͳ໰୊ΛੜΈग़ͯ͠͸͍͚ͳ͍ w ҆қͳػցֶशͷཚ༻͸ɺඞͣޙͷෛ࠴ʹܨ͕Δ w ෳ਺ͷ໰୊ΛҰ౓ʹղ͜͏ͱͯ͠͸͍͚ͳ͍ w ೋ㙽௥͏΋ͷ͸Ұ㙽Λ΋ಘͣ w ྫɿίετ࡟ݮͭͭ͠རӹ΋࠷େԽ͢Δਓ޻஌ೳ"*

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ϋϯϚʔΛ࣋ͭਓʹ͸ɺ͢΂͕ͯఝʹݟ͑Δ *GBMMZPVIBWFJTBIBNNFS FWFSZUIJOHMPPLTMJLFBOBJM ΞϒϥϋϜɾϚζϩʔ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ظ଴஋Λௐ੔͢Δ w ਫ਼౓΁ͷظ଴ w ਓ͕ؒ׬ᘳʹͰ͖ͳ͍Α͏ʹɺػց͕ͷਫ਼౓Λग़͢͜ͱ͸ෆՄೳ w ਓؒʹ͚ۙͮΔ͜ͱ͸Մೳ w
੒Ռ΁ͷظ଴ w ࠷ۙͷ"*͸ࣗಈͰYY͕Ͱ͖Δˠਓؒͷ࢓ࣄ͕ແ͘ͳΔʜʜʁ w কདྷతͳՄೳੑ΋ؚΊͭͭɺݱ࣌఺ͰͷՄೳෆՄೳΛ໌֬ʹ͢Δ w ධՁͷํ๏Λ͋Β͔͡ΊܾΊ͓ͯ͘ w ໨ඪͷఆྔԽΛͯ͠ɺؒҧͬͨํ޲ͷ౒ྗΛ͠ͳ͍ w Ϗδωεଆͷਓؒͱٕज़ऀ͕ɺڞ௨ͨ͠਺ࣈͰٞ࿦͢Δ͜ͱ͕େࣄ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE w ਫ਼౓໘Ͱܧଓతͳվળ͕ඞཁ w ࠷ॳ͔Βਫ਼౓͕ཧ૝஋·Ͱ্͕ΔΘ͚Ͱ͸ͳ͍ w ਫ਼౓Λ্͍͛ͯ͘஍ಓͳ࡞ۀ͕ඞཁ w ਫ਼౓ΛৗʹධՁ͠ϞχλϦϯά͠ଓ͚Δඞཁ͕͋Δ
w ௕ظత൐૸ͷϙδςΟϒͳଆ໘ w σʔλ͕஝ੵ͍ͯ͘͜͠ͱͰਫ਼౓͸޲্͍ͯ͘͠ w ʮ໦͜ΓͷδϨϯϚʯʹؕΒͳ͍ w ໨ͷલͷ՝୊΋େࣄ͕ͩɺઌΛݟӽͯ͠౤ࢿΛ͢Δͷ΋େࣄ ௕ظతͳ൐૸Λ֮ޛ͢Δ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ػցֶश෦෼ΛγεςϜຊମ͔Β෼཭ͤ͞Δ w ؔ܎ੑΛૄʹ͢Δ w ద͢Δ։ൃελΠϧ͕ͦ΋ͦ΋ҧ͏ w ΢ΥʔλʔϑΥʔϧܕ։ൃWTΞδϟΠϧܕ։ൃ w
ຊମγεςϜͷ։ൃ͕ऴྃͯ͠΋ػցֶश෦෼ͷ؅ཧ͸ଓ͘ w ಠཱͯ͠։ൃ w ΠϯϓοτΞ΢τϓοτΛܾΊ͓͚ͯ͹ɺத਎ͷਫ਼౓޲্͸ຊମγεςϜʹӨڹ ͞Εͣʹ࣮ࢪͰ͖Δ w ϚΠΫϩαʔϏεతͳߟ͑ํ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE γϯϓϧͳํ๏͔Βࢼ͍ͯ͘͠ w ؆୯ͳํ๏͔Βࢼͯ͠ɺঃʑʹෳࡶͳख๏ʹਂ۷Δ w ͢΂ͯʹ͓͚ΔධՁͷج४Λ࡞Δ w ج४͕ແ͚Ε͹ਫ਼౓΍඼࣭ͷྑ͠ѱ͠Λ൑அͰ͖ͳ͍ w
ෳࡶͳ͜ͱʹ࣌ؒΛ͔͚ͯ͠·͏͜ͱʹΑΔख໭ΓΛແ͘͢ w ʮ͋Δ໰୊Λղ͘ͱ͖ɺͦͷ໰୊ΑΓ΋೉͍͠໰୊Λ్தͰղ͍ͯ͸ͳΒͳ͍ʯ CZ7BQOJLT1SJODJQMF

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ݁ՌΛਖ਼͘͠ଌΔ w ߏஙͨ͠ػցֶश͕Ϗδωεͱͯ͠Ձ஋͕͋Δ͔Ͳ͏͔ w ΍ͬͯҙຯ͕͋Δͷʁ͋ͬͨͷʁʹ౴͑Δ w ࢪࡦͷલޙͰͲͷ,1*͕Ͳ͏มΘͬͨͷ͔Λଌఆ w
"#ςετʹΑΔ֬ೝ΋ॏཁ w ϩδοΫ͕ϒϥοΫϘοΫεͰ͋Δ͜ͱͷෆ҆ΛऔΓআ͘ w ਓؒͷײ֮ͱػցͷਫ਼౓͕Ұக͠ͳ͍͜ͱ͕͋Δ w ྫɿϨίϝϯυɾਪનγεςϜ

$PQZSJHIU3FDSVJU5FDIOPMPHJFT$P -UE"MM3JHIUT3FTFSWFE ·ͱΊ w αʔϏεΛ࢖͏ਓʹ࢖ΘΕΔ΋ͷΛ૑Δ w Ϗδωεͱͯ͠Ձ஋ͷ͋ΔϞϊΛػցֶश"*Ͱ࣮ݱ͢Δ w ৽͍ٕ͠ज़Λ͍ͪૣ͘ಋೖ͠ɺϦΫϧʔτ಺෦ʹఏڙ͍ͯ͠Δ
w αʔϏεΛ࡞Δਓʹ࢖ΘΕΔ΋ͷΛ૑Δ w ෳࡶԽ͍ͯ͘͠ػցֶश"*Λ࢖ͬͨϓϩμΫτͷ։ൃΛՃ଎͍ͯ͘͠ w ΦʔϓϯΠϊϕʔγϣϯͱͯ͠ߩݙͭͭ͠ɺϑΟʔυόοΫΛಘΔ w ػցֶशͷ"1*܈"35ͷ֎෦ެ։ w ͥͻ࢖ͬͯΈͯԼ͍͞ʂ<"35>Ͱݕࡧʂ

リクルートグループの現場事例から見るAI/ディープラーニング技術のビジネス活用と“A3RT”外...

リクルートグループの現場事例から見るAI/ディープラーニング技術のビジネス活用と“A3RT”外部公開の裏側

More Decks by Recruit Technologies

Other Decks in Technology

Featured

Transcript