Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
論文紹介 Balancing Relevance and Discovery to Inspi...
Search
Takashi Nishibayashi
October 17, 2020
Research
0
730
論文紹介 Balancing Relevance and Discovery to Inspire Customers in the IKEA App
RecSys2020論文読み会の発表資料です
https://connpass.com/event/189192/
Takashi Nishibayashi
October 17, 2020
Tweet
Share
More Decks by Takashi Nishibayashi
See All by Takashi Nishibayashi
診断前の病歴テキストを対象としたLLMによるエンティティリンキング精度検証
hagino3000
1
120
論文紹介 Improving Medical Reasoning through Retrieval and Self-Reflection with Retrieval-Augmented Large Language Models
hagino3000
0
860
論文紹介 Audience Size Forecasting Fast and Smart Budget Planning for Media Buyers
hagino3000
0
240
論文紹介 Towards a Fair Marketplace: Counterfactual Evaluation of the trade-off between Relevance, Fairness & Satisfaction in Recommendation Systems
hagino3000
1
630
論文紹介 Budget Management Strategies in Repeated Auctions (公開版)
hagino3000
2
290
論文紹介 A Request-level Guaranteed Delivery Advertising Planning: Forecasting and Allocation
hagino3000
1
120
論文紹介 Online Experimentation with Surrogate Metrics Guidelines and a Case Study
hagino3000
1
340
論文紹介 Bidding Machine: Learning to Bid for Directly Optimizing Profits in Display Advertising
hagino3000
1
210
不確実性と上手く付き合う意思決定の手法
hagino3000
19
15k
Other Decks in Research
See All in Research
「どう育てるか」より「どう働きたいか」〜スクラムマスターの最初の一歩〜
hirakawa51
0
820
経済学と機械学習:因果推論と密度比推定を中心に
masakat0
0
140
SSII2025 [SS2] 横浜DeNAベイスターズの躍進を支えたAIプロダクト
ssii
PRO
7
3.9k
Self-supervised audiovisual representation learning for remote sensing data
satai
3
260
なめらかなシステムと運用維持の終わらぬ未来 / dicomo2025_coherently_fittable_system
monochromegane
0
2.3k
時系列データに対する解釈可能な 決定木クラスタリング
mickey_kubo
2
900
2025年度人工知能学会全国大会チュートリアル講演「深層基盤モデルの数理」
taiji_suzuki
25
18k
数理最適化に基づく制御
mickey_kubo
6
720
Computational OT #4 - Gradient flow and diffusion models
gpeyre
0
370
【輪講資料】Moshi: a speech-text foundation model for real-time dialogue
hpprc
3
620
Streamlit 総合解説 ~ PythonistaのためのWebアプリ開発 ~
mickey_kubo
1
1.4k
GeoCLIP: Clip-Inspired Alignment between Locations and Images for Effective Worldwide Geo-localization
satai
3
300
Featured
See All Featured
Templates, Plugins, & Blocks: Oh My! Creating the theme that thinks of everything
marktimemedia
31
2.5k
Faster Mobile Websites
deanohume
309
31k
GraphQLとの向き合い方2022年版
quramy
49
14k
The Invisible Side of Design
smashingmag
301
51k
The Cost Of JavaScript in 2023
addyosmani
53
8.8k
Building an army of robots
kneath
306
46k
Designing for humans not robots
tammielis
253
25k
Building a Scalable Design System with Sketch
lauravandoore
462
33k
Creating an realtime collaboration tool: Agile Flush - .NET Oxford
marcduiker
31
2.2k
Optimising Largest Contentful Paint
csswizardry
37
3.4k
Understanding Cognitive Biases in Performance Measurement
bluesmoon
29
1.8k
Site-Speed That Sticks
csswizardry
10
780
Transcript
จհ #BMBODJOH3FMFWBODFBOE%JTDPWFSZ UP*OTQJSF$VTUPNFSTJOUIF*,&""QQ 3FD4ZTจಡΈձ ྛ !IBHJOP
"CPVUNF ✦ 4PGUXBSF&OHJOFFSBU70:"(&(3061 ✦ ࠂ৴ϓϩμΫτͷ։ൃΛ͍ͯ͠·͢ ✦ 3FD4ZTॳࢀՃ ✦ 5XJUUFS!IBHJOP
հ͢Δจ
༰ ✦ *,&"ΞϓϦͷ*OTQJSBUJPOBM'FFEʹ͓͚ΔϦίϝϯυํࡦʹ#BOEJU 'FFECBDLϩάͬͨϙϦγʔֶशΞϓϩʔνΛద༻ͨ͠ ✦ Ϣʔβʔ͕ΠϯεϐϨʔγϣϯΛಘΒΕΔ༷ʹDPVOUFSGBDUVBMSJTL NJOJNJ[BUJPOQSJODJQMF<>ʹج͍ͮͯํࡦΛֶशͨ͠ ✦ "#ςετͷ݁ՌɺڠௐϑΟϧλϦϯάʹΑΔํࡦͱൺֱͯ͠ΫϦοΫ ্͕ঢͨ͠
✦ ϖʔύʔʹଛࣦ࣮ؔݧ݁Ռ΄ͱΜͲॻ͍ͯͳ͔ͬͨͷͰɺ͜ͷൃ දޱ಄ൃද༰Λͬͯิ͍ͯ͠·͢
*OTQJSBUJPOBM'FFE ✦ Ϣʔβʔ༷ʑͳ෦ͷλΠϓʹ͋ΘͤͯՈ۩͕ ஔ͞ΕͨΠϝʔδ͕ӾཡͰ͖Δ ✦ ͜ͷϑΟʔυϢʔβʔʹؔ࿈͕͋ΓΠϯεϐ ϨʔγϣϯΛ༩͑Δͷʹ͍ͨ͠ɻڻ͖ͷཁૉ͕ ٻΊΒΕΔɻ 3FMFWBODFBOE%JTDPWFSZUP*OTQJSF
✦ ؔ࿈ੑʹಛԽͨ͠Ϧίϝϯυػೳطʹ͋Δ
$POUFYUVBMCBOEJUTXJUICBUDIMFBSOJOHGSPN MPHHFECBOEJUGFFECBDL ✦ Ͳͷը૾Λදࣔ͢Δ͔ ✦ $POUFYUVBM#BOEJUTͰܾΊΔ ✦ $POUFYUVBM#BOEJUTͷߦಈબϙϦγʔͷֶश ✦ όονͰΔ
㱠ΦϯϥΠϯֶश ✦ #BOEJU'FFECBDLϩάΛͬͯ܇࿅ ✦ CBTFEPOUIFQSJODJQMFPGDPVOUFSGBDUVBMSJTLNJOJNJ[BUJPO<> ✦ $PVOUFSGBDUVBM-FBSOJOH
✦ ߦಈ ✦ ը૾Λબͯ͠දࣔ͢Δࣄ ✦ ใु ✦ දࣔͨ͠ը૾͕ΫϦοΫ͞ΕΔ͔Ͳ͏͔㱨\ ^ ✦
ίϯςΩετ ✦ ΞϓϦ্ͷϢʔβʔࣗͷաڈͷ;Δ·͍ *OTQJSBUJPOBM'FFEͷ#BOEJUઃఆ
$POUFYUVBMCBOEJUT ܁Γฦ͠ҙࢥܾఆʹ͓͍ͯྦྷੵใुͷ࠷େԽΛૂ͏ํࡦͷͳ͔Ͱ ϥϯυຖͷίϯςΩετใΛར༻ͯ͠ٻΊͨείΞʹैͬͯߦಈΛબ͢ Δͷɻ؍ଌͨ͠ใुΛͬͯείΞϦϯάϞσϧΛߋ৽͍ͯ͘͠ɻ είΞϦϯάϞσϧʹઢܗϞσϧΛ࠾༻ͨ͠-JO6$#<>ͳͲɺ༷ʑͳํࡦ͕ఏҊ͞Ε͍ͯΔ
ࢀߟ-JO6$#<> ϥϯυUʹ͓͚ΔߦಈBͷ είΞใुͷظ ඪ४ภࠩºЋ είΞ͕࠷େͷߦಈΛ࣮ߦ ύϥϝʔλߋ৽
#BUDIMFBSOJOHGSPNMPHHFECBOEJUGFFECBDL ✦ CBOEJUGFFECBDLϩάΛֶͬͨश ✦ աڈͷߦಈબϙϦγʔʹΑΔόΠΞεͷิਖ਼͕ඞཁ ✦ ਪનγεςϜͷϩάجຊతʹCBOEJUGFFECBDL ✦ ΦϯϥΠϯֶशͰͳ͍ཧ༝ಛʹઆ໌͕ແ͔͕ͬͨ ✦
ϦΫΤετྔ͕ଟ͍αʔϏεͰόϯσΟοτΞϧΰϦζϜΛ͏߹ จͷखଓ͖௨ΓʹύϥϝʔλΛஞ࣍ߋ৽͢Δέʔεগͳ͍ͱࢥ͏ ✦ ΦϯϥΠϯֶशӡ༻͕େม
ิ#BOEJU'FFECBDLϩάΛֶͬͨश ✦ #BOEJU'FFECBDLϩάΛར༻ͨ͠৽͍͠ߦಈબϙϦγʔͷੑೳධՁΛߦͳ͏ ख๏ଘࡏ͢Δ ˠ0⒎1PMJDZ&WBMVBUJPO *OWFSTF1SPQFOTJUZ4DPSJOH %PVCMZ3PCVTU ʜ ✦
ੑೳධՁ͕࠷େʹͳΔ༷ʹߦಈબϙϦγʔΛֶश͢Εྑ͍ ✦ ߦಈΛBɺίϯςΩετΛYɺใुΛSͱͯ͠*14ͰධՁ͢Δ߹ КOFXͷ*14$PVOUFSGBDUVBM&TUJNBUPS
ߦಈબϙϦγʔͷֶश ϙϦγʔК͕ίϯςΩετYʹରͯ͠ߦಈZΛબͿ֬ είΞ ΛК ZcY ใुΛЎͱ͓͘ɻաڈͷϙϦγʔКͷείΞͱ؍ଌͨ͠ใुΛར༻ͯ͠ɺ৽ͨͳ ϙϦγʔКВΛֶश͢Δ ޱ಄ൃදεϥΠυ͔Βഈआ ͜ͷ··Ͱࢄͷ͕ग़ΔͷͰɺ<>Λࢀߟʹ͍͔ͭ͘ͷΛऔΓ͍Ε͍ͯΔͱͷઆ
໌͕͋ͬͨ
ଛࣦؔͷײతͳղऍ ใु͕ಘΒΕͨߦಈͰಛʹաڈͷϙϦγʔͷείΞ͕͍ߦಈͷείΞ্͕ ͕Εϩε͕Լ͕Δɻͭ·Γ͋·Γબ͠ͳ͔͕ͬͨΫϦοΫ͕ಘΒΕͨߦಈΛଟ͘ બͿ༷ʹֶश͢Δɻ ޱ಄ൃදεϥΠυ͔Βഈआ
ଛࣦؔͲ͔͜Βདྷͨͷ͔ ޱ಄ൃදεϥΠυͷࣜ3FGFSFODFʹ͋Δ 4XBNJOBUIBOΒ$PVOUFSGBDUVBMSJTL NJOJNJ[BUJPO-FBSOJOHGSPNMPHHFECBOEJU GFFECBDLz *$.- <>ʹ͓͚ΔఏҊख๏ ͷಋग़ͷং൫ʹ͋Δ*14ϕʔεͷࣜɻ ͳͷͰ࣮ࡍʹ͍ͬͯΔ
ͷ<>ͷఏҊख๏ͷࣜ ͩͱࢥΘΕΔ
$PVOUFSGBDUVBMSJTLNJOJNJ[BUJPO-FBSOJOHGSPNMPHHFE CBOEJUGFFECBDL *$.- <> ✦ *14ͷࢄ༝དྷͷΤϥʔΛόϯυ͢ΔVOCJBTFEFTUJNBUPSͰ͋Δ $3.$PVOUFSGBDUVBM3JTL.JOJNJ[BUJPOΛఏҊɺࢄΛਖ਼ଇԽ߲ʹ ✦ $3.Λֶश͢Δ܇࿅ΞϧΰϦζϜ10&.ͷఏҊ
ิ%PVCMZSPCVTUNFUIPEGPSDPVOUFSGBDUVBM MFBSOJOH Yuan, Bowen, et al. "Improving ad click prediction
by considering non-displayed events." Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 2019.
ΦϯϥΠϯධՁ ✦ ධՁࢦඪ$53 ✦ ͷ্͕֬ೝͰ͖ͨ ✦ ڠௐϑΟϧλϦϯάϕʔεͷख๏ͱൺֱ ✦ ڠௐϑΟϧλϦϯά3FMFWBODFॏࢹ ✦
ॳͷతୡͰ͖ͨ ✦ આ໌ແ͔͕ͬͨʮࠓ·Ͱʹਪન͠ͳ͔ͬͨΞΠςϜΛଟ͘ਪન͢ΔࣄͰΫ ϦοΫ͕૿͑ͨʯˠ*OTQJSBUJPOΛ༩͑Δࣄ͕Ͱ͖ͨͱղऍͰ͖ͦ͏ ✦ ΫϦοΫ͕ݮ͍ͬͯͳ͍ˠϏδωεࢦඪΛᆝଛ͍ͯ͠ͳ͍
ͷ࣭ٙԠͷൈਮ 2%PVCMZ3PCVTUͰͳ͘*14ʹͨ͠ͷԿނ ಛʹΦϑϥΠϯධՁͰࢄ͕ʹͳΒͳ͔͔ͬͨ "%PVCMZ3PCVTUߟ͑ͳ͔ͬͨɻ͔͠͠ࢄΛ͑ΔͨΊʹεί ΞͷΫϦοϐϯάΛߦͳͬͨ 2"#ςετΛि͚ؒͩΒͤͨ͜ͱͰɺϢʔβʔ͕׳Ε͠Μͩ " ͱҧͬ
͍ͯͨͨΊ # ͷ$53͕ߴ͘ͳͬͨՄೳੑ͋Δ͔ "ͦͷޙҰ؏ͯ͠ߴ͍$53Λ͍ࣔͯ͠Δ͔ΒɺͦΕແ͍ͱߟ͍͑ͯΔ 2୳ࡧΛߦͳ͏ࣄͰΫϦοΫͷݮগΈΒΕͳ͔͔ͬͨ "શ͘ٯͰ૿Ճͨ͠
ࢀߟจݙ <>5ÓUI #BMÂ[T 4BOEIZB4BDIJEBOBOEBO BOE&NJM4+SHFOTFO#BMBODJOH3FMFWBODFBOE %JTDPWFSZUP*OTQJSF$VTUPNFSTJOUIF*,&""QQ'PVSUFFOUI"$.$POGFSFODFPO 3FDPNNFOEFS4ZTUFNT <>-J -JIPOH FUBM"DPOUFYUVBMCBOEJUBQQSPBDIUPQFSTPOBMJ[FEOFXTBSUJDMF
SFDPNNFOEBUJPO1SPDFFEJOHTPGUIFUIJOUFSOBUJPOBMDPOGFSFODFPO8PSMEXJEFXFC <>4XBNJOBUIBO "EJUI BOE5IPSTUFO+PBDIJNT$PVOUFSGBDUVBMSJTLNJOJNJ[BUJPO-FBSOJOH GSPNMPHHFECBOEJUGFFECBDL*OUFSOBUJPOBM$POGFSFODFPO.BDIJOF-FBSOJOH <>:VBO #PXFO FUBM*NQSPWJOHBEDMJDLQSFEJDUJPOCZDPOTJEFSJOHOPOEJTQMBZFE FWFOUT1SPDFFEJOHTPGUIFUI"$.*OUFSOBUJPOBM$POGFSFODFPO*OGPSNBUJPOBOE ,OPXMFEHF.BOBHFNFOU