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
20180414_WSDM2018_reading_YoheiKIKUTA
Search
yoppe
April 12, 2018
Science
0
690
20180414_WSDM2018_reading_YoheiKIKUTA
HP:
https://atnd.org/events/95510
yoppe
April 12, 2018
Tweet
Share
More Decks by yoppe
See All by yoppe
20211023_recsys2021_paper_reading_YoheiKikuta
diracdiego
2
450
20201121_oldpaperreading_computing_machinery_and_intelligence
diracdiego
0
150
20200906_ACL2020_metric_for_ordinal_classification_YoheiKikuta
diracdiego
1
1.2k
20191102_ACL2019_adversarial_examples_in_NLP_YoheiKIKUTA
diracdiego
2
1.4k
20190223_nlpaperchallenge_CV_4.3to5.5
diracdiego
2
760
20180701_CVPR2018_reading_YoheiKIKUTA
diracdiego
3
1.1k
20180306_NIPS2017_DeepLearning
diracdiego
4
5.8k
20180215_MLKitchen7_YoheiKIKUTA
diracdiego
0
380
20180210_Cookpad_TechConf2018_YoheiKIKUTA
diracdiego
5
1.1k
Other Decks in Science
See All in Science
構造設計のための3D生成AI-最新の取り組みと今後の展開-
kojinishiguchi
0
490
非同期コミュニケーションの構造 -チャットツールを用いた組織における情報の流れの設計について-
koisono
0
130
Презентация программы бакалавриата СПбГУ "Искусственный интеллект и наука о данных"
dscs
0
690
Iniciativas independentes de divulgação científica: o caso do Movimento #CiteMulheresNegras
taisso
0
130
統計的因果探索の方法
sshimizu2006
1
1.1k
Introduction to Graph Neural Networks
joisino
PRO
4
2k
Celebrate UTIG: Staff and Student Awards 2024
utig
0
430
機械学習による確率推定とカリブレーション/probabilistic-calibration-on-classification-model
ktgrstsh
2
210
(2024) Livres, Femmes et Math
mansuy
0
100
As We May Interact: Challenges and Opportunities for Next-Generation Human-Information Interaction
signer
PRO
0
130
Презентация программы магистратуры СПбГУ "Искусственный интеллект и наука о данных"
dscs
0
370
(Forkwell Library #48)『詳解 インシデントレスポンス』で学び倒すブルーチーム技術
scientia
2
1.3k
Featured
See All Featured
Thoughts on Productivity
jonyablonski
67
4.3k
Documentation Writing (for coders)
carmenintech
65
4.4k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
191
16k
Build The Right Thing And Hit Your Dates
maggiecrowley
32
2.4k
I Don’t Have Time: Getting Over the Fear to Launch Your Podcast
jcasabona
27
1.9k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
664
120k
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
4
290
Building Better People: How to give real-time feedback that sticks.
wjessup
363
19k
The Pragmatic Product Professional
lauravandoore
31
6.3k
Building a Scalable Design System with Sketch
lauravandoore
459
33k
Making the Leap to Tech Lead
cromwellryan
132
8.9k
ピンチをチャンスに:未来をつくるプロダクトロードマップ #pmconf2020
aki_iinuma
107
49k
Transcript
Why People Search for Images using Web Search Engines WSDM
2018 จಡΈձ 20180414 ٠ా ངฏ (@yohei_kikuta) Event URL: https://atnd.org/events/95510, paper: https://arxiv.org/abs/1711.09559
·ͱΊ 1. text base ͷΣϒը૾ݕࡧͷҙਤྨՄೳ͔ʁ → YES. 3ͭʹྨ: Entertain, Explore/Learn,
Locate/Acquire 2. औಘՄೳͳಛྔ͔ΒҙਤΛผͰ͖Δ͔ʁ → YES. ཹ࣌ؒϚεδΣενϟ 3. ηογϣϯॳظͰݕࡧҙਤΛ༧ଌͰ͖Δ͔ → MAYBE. ಛྔΛͬͯϞσϧΛ࡞ͯ͠Ұఆͷੑೳ 2
എܠ 3
ݕࡧͷҙਤΛΓ͍ͨ Ϣʔβͷݕࡧߦಈͷཪʹ͋ΔҙਤΛΔ͜ͱॏཁ → Ϣʔβͷຬ্ʢsuggestion, recommendation, ...ʣ Σϒݕࡧͷݚڀͳ͞Ε͖͕ͯͨɺը૾ݕࡧʹؔͯ͠ݶఆత → ΫΤϦϕʔε →
͔͠͠ը૾ݕࡧͷΫΤϦ͘ͳΓ͕ͪͰෆ࣮֬ੑ͕େ͖͍ ຊจͰηογϣϯใΛѻͬͯը૾ݕࡧͷҙਤΛݚڀ 4
ຊจʹ͓͚ΔϦαʔνΫΤενϣϯ 1. text base ͷΣϒը૾ݕࡧͷҙਤྨՄೳ͔ʁ 2. औಘՄೳͳಛྔ͔ΒҙਤΛผͰ͖Δ͔ʁ 3. ηογϣϯॳظͰݕࡧҙਤΛ༧ଌͰ͖Δ͔ 5
ઌߦݚڀ 6
Σϒݕࡧʹ͓͚Δҙਤͷ taxonomy A taxonomy of web search (2002) ͰҙਤΛ3ͭʹྨ 1.
Navigational: ಛఆͷαΠτ౸ୡ 2. Informational: ใͷऔಘ 3. Transactional: ΣϒΛഔհͱͨ͠׆ಈ Ref: https://dl.acm.org/citation.cfm?id=792552 7
Σϒݕࡧʹ͓͚Δҙਤͷ taxonomy Task Behaviors During Web Search: The Difficulty of
Assigning Labels (2009) ͰݕࡧλεΫΛ7ͭʹྨ » Navigate, Find-Simple, Find-Complex, Locate/Acquire, Explore/Learn, Play, Meta ຊจ͜ͷઌߦݚڀΛ౿ऻͭͭ͠ը૾ݕࡧʹൃలͤͨ͞ͷɺͱ͍͏ ৭߹͍͕ڧ͍ Ref: http://ieeexplore.ieee.org/document/4755491/ 8
ը૾ݕࡧͷҙਤΛྨ 9
Ξϓϩʔν σʔλΛूΊͯͦΕΛجʹ3ਓͷΣϒݚڀऀ͕ྨ » ϢʔβͷΞϯέʔτσʔλ » ੑผใͳͲΛऔಘ » ࠷ۙͷݕࡧʹؔ͢ΔৄࡉʢಈػͳͲʣɺ༻ͨ͠ΫΤϦ » దͳճΛͨ͠211ਓ͕ର
10
Ξϓϩʔν σʔλΛूΊͯͦΕΛجʹ3ਓͷΣϒݚڀऀ͕ྨ » ϩάσʔλ » https://www.sogou.com/ ͷϩάσʔλ » 30Ҏʹ࿈ଓతͳΫΤϦΛ༩͍͑ͯΔ475ηογϣϯʢআ͘Ξμ ϧτʣ
11
Ξϓϩʔν σʔλΛूΊͯͦΕΛجʹ3ਓͷΣϒݚڀऀ͕ྨ » ϩάσʔλʢlength ΫΤϦʣ Ref: https://arxiv.org/abs/1711.09559 12
࡞ͨ͠அج४ 1. Ϣʔβͷݕࡧߦಈ໌֬ͳతʹґΔͷ͔ʁ 2. ޙͷར༻ͷͨΊʹը૾Λμϯϩʔυ͢Δඞཁ͕͋Δ͔ʁ 13
3ͭͷݕࡧҙਤ 1. Explore/Learn (1-yes, 2-no) ྫʣΰϦϥͱϘϊϘͷݟͨͷҧ͍ΛνΣοΫ 2. Locate/Acquire (1-yes, 2-yes)
ྫʣϨϙʔτ࡞Ͱ͏ΰϦϥͷը૾Λ୳ͯ͠μϯϩʔυ 3. Entertain (1-no, 2-yes or no) ྫʣΰϦϥͷ໘നը૾ΛோΊΔ 14
3ͭͷݕࡧҙਤʢྫʣ Ref: https://arxiv.org/abs/1711.09559 15
ଥੑͷݕূʢ3ਓͷେֶӃੜʹΑΔҙਤྨʣ » ϢʔβͷΞϯέʔτσʔλ Fleiss' kappa: 0.673 Explore/Learn: 27%, Locate/Acquire: 66%,
Entertain: 7% » ϩάσʔλʢΫΤϦͷΈΛ༻ʣ Fleiss' kappa: 0.375 Explore/Learn: 56%, Locate/Acquire: 39%, Entertain: 5% ͏·͚͘Εͦ͏͕ͩΫΤϦͷΈͰҙਤΛΉͷ͍͠ 16
औಘՄೳͳಛྔͰҙਤΛผ 17
35ਓͷֶ෦ੜʹΑΔ12ݸͷը૾ݕࡧλεΫ ྫʣPCͷഎܠΛ੨ۭͱͷը૾ʹมߋʢLocate/Acquireʣ ͦͷࡍʹҎԼͷಛྔΛऔಘ Ref: https://arxiv.org/abs/1711.09559 18
ҙਤʹΑͬͯ༗ҙͳ͕ࠩग़ΔͷͰผՄೳ ఀཹ࣌ؒ E/L ͕ଟ͍ɺϚεΫϦοΫ E/L < L/A < EɺͳͲ ʢৄࡉจΛࢀরʣ
Ref: https://arxiv.org/abs/1711.09559 19
ηογϣϯॳظͰͷҙਤͷ༧ଌ 20
ઃఆ ηογϣϯॳظͱʮ࠷ॳͷϚεεΫϩʔϧ͕͋Δ·Ͱʯ ༧ଌͰ͏ feature ͱͯ͠ҎԼͷҙ - ΫϦοΫͱ࠷ॳͷϚεΦʔόʔ࣌ؒΘͳ͍ - ΫΤϦϕʔεͰΓ͍ͨͷͰ query
reformulation Θͳ͍ ֶ෦ੜʹղ͔ͤͨը૾ݕࡧλεΫʹରͯ͠ GBDT Ͱ 10-fold CV 21
༧ଌੑೳߴ͘ͳ͍͕ෆՄೳͰͳͦ͞͏ Baseline majority ʹશ෦دͤΔͱ͍͏ͷ Ref: https://arxiv.org/abs/1711.09559 22
·ͱΊͱॴײ 23
·ͱΊʢ࠶ܝʣ 1. text base ͷΣϒը૾ݕࡧͷҙਤྨՄೳ͔ʁ → YES. 3ͭʹྨ: Entertain, Explore/Learn,
Locate/Acquire 2. औಘՄೳͳಛྔ͔ΒҙਤΛผͰ͖Δ͔ʁ → YES. ཹ࣌ؒϚεδΣενϟ 3. ηογϣϯॳظͰݕࡧҙਤΛ༧ଌͰ͖Δ͔ → MAYBE. ಛྔΛͬͯϞσϧΛ࡞ͯ͠Ұఆͷੑೳ 24
ॴײ » γϯϓϧͳج४ͰྨΛ͍ͯ͠Δͱ͍͏ͷྑ͍ » ৽ͱ͍͏Θ͚Ͱͳ͍͕ҰͭҰ͔ͭͬ͠Γௐ͍ͯΔ » ࣮αʔϏεͷԠ༻ʹҰาඈ༂͕ඞཁͦ͏ʢ༧ଌੑೳͳͲʣ » ٱʑʹࣜΛશવΘͳ͍จΛಡΜͰ৽ͩͬͨ 25