7. 14 Workshops 7
W1: TempWeb
W2: MAISoN 2020
W3: SocialNLP 2020
W4: ECNLP 2
W5: DSSGW
W6: SIdEWayS
W7: IID
W8: CyberSafety 2020
W9: DecentWeb
W10: DL4G 2020
W11: LocWeb2020
W12: II FATES
W13: Wiki Workshop2020
W14: WebAndTheCity
Jiawei Han
University of Illinois Urbana-Champaign
On What Kinds of Networks
Do We Need Deep Learning Most?
Jure Leskovec
Stanford University
Advancements in Graph Neural
Networks: PGNNs, Pretraining, & OGB
W10: DL4G 2020 の keynotes
8. 13 Tutorials 8
T1: Wikimedia Public (Research) Resources
T2: Entity Summarization in Knowledge Graphs: Algorithms Evaluation and Applications
T3: Fairness and Bias in Peer Review and other Sociotechnical Intelligent Systems
T4: Explainable AI in Industry: Practical Challenges and Lessons Learned
T5: Learning Graph Neural Networks with Deep Graph Library
T6: Constructing Knowledge Graph for Social Networks in A Deep and Holistic Way
T7: Mining signed networks: theory and applications
T8: Unbiased Learning to Rank: Counterfactual and Online Approaches
T9: Deep Transfer Learning for Search and Recommendation
T10: Methods of Corporate Surveillance: a Primer on Experimental Transparency Research
T11: Challenges Best Practices and Pitfalls in Evaluating Results of Online Controlled Experiments
T12: Forecasting Big Time Series: Theory and Practice
T13: Responsible Recommendation and Search Systems
11. Special speaker 11
Audrey Tang
Minister without Portfolio, Executive Yuan
Digital Social Innovation: Taiwan Can Help
12. 3 Keynotes 12
Yolanda Gil
University of Southern California
Embedding the Scientific Record on the Web:
Towards Automating Scientific Discoveries
Sir Nigel Shadbolt
University of Oxford
Architectures for Autonomy:
Towards an Equitable Web of Data in the Age of AI
Wei-Ying Ma
ByteDance
Democratizing Content Creation and
Dissemination through AI Technology
Webに知識を埋め込む
ユーザデータの公平な取り扱い
コンテンツ創作を支援するAI技術
14. 投稿論文の各種統計量 14
12 research tracks(投稿時にトラックを指定)
2180 abstracts submitted
1600+ abstracts converted to paper submissions
~1500 eligible for regular review after withdraw and desk rejection
317 papers accepted (219 full papers; 98 short papers)
20. 12種類のTrackと発表件数 20
Track 件数
Web Mining and Content Analysis 61
User Modeling 52
Social Network Analysis and Graph Algorithms 51
Semantics and Knowledge 25
Web and Society 22
Security, Privacy, and Trust 21
Web of Things, Ubiquitous, and Mobile Computing 18
Search 17
Health on the Web 15
Economics, Monetization, and Online Markets 13
Intelligent Systems and Infrastructure 11
Crowdsourcing and Human Computation 10
10種類のTrackから1本ずつ論文を紹介
21. Best Paper
Track: User Modeling
Open Intent Extraction from Natural Language
Interactions
Nikhita Vedula (The Ohio State University)
Nedim Lipka (Adobe)
Pranav Maneriker (The Ohio State University)
Srinivasan Parthasarathy (The Ohio State University)
22. 目的 22
ユーザの発話内容からユーザの意図を抽出することが目的
意図 = action + object
既定の意図カテゴリへの分類でなく任意の意図を抽出するのが最大の特徴
I would like to reserve a seat
and request a special meal on my flight.
(action, object) = (reserve, seat), (request, special meal)
23. 手法概要 23
入力文中の各単語に対して Action・Object・None のいずれかをタグ付け
I would like to reserve a seat …
None None None None Action None Object
敵対的学習で頑健性向上
マルチヘッドを採用して
全単語間の複数の関係を考慮
27. Best Student Paper
Track: Web of Things, Ubiquitous, and Mobile Computing
Mobile App Squatting
Yangyu Hu (BUPT), Haoyu Wang (Beijing University of Posts and Telecommunications),
Ren He (Beijing University of Posts and Telecommunications), Li Li (Monash University),
Gareth Tyson (Queen Mary University of London), Ignacio Castro (Queen Mary University of London),
Yao Guo (Peking University), Lei Wu (Zhejiang University),
Guoai Xu (Beijing University of Posts and Telecommunications)
33. Track: Web and Society
Don’t Let Me Be Misunderstood: Comparing
Intentions and Perceptions in Online Discussions
Jonathan P. Chang (Cornell University)
Justin Cheng (Facebook)
Cristian Danescu-Niculescu-Mizil (Cornell University)
39. Track: Health on the Web
The Automated Copywriter:
Algorithmic Rephrasing of Health-Related
Advertisements to Improve their Performance
Brit Youngmann (Microsoft Research)
Elad Yom-Tov (Microsoft Research)
Ran Gilad-Bachrach (Microsoft Research)
Danny Karmon (Microsoft Healthcare NExT)
40. 目的 40
Singling Out Shingles Vaccine –
13 Health Facts. Check out 13 health facts
about shingles on ActiveBeat right now.
広告を出す会社が
考えた広告
提案手法
提案手法により
改善された広告
Shingles Vaccine – everything you need
to know. Discover 13 fact on shingles on
ActiveBeat. Get expert advice now!
SERPに表示する医療関係の広告の文章をより魅力的に変換してクリック率改善
アプローチ:元の広告の文章を魅力的な文章に翻訳
41. 提案手法の流れ1 41
Normalization
Candidate generation
Preprocessing
Normalization
医療関連の単語を記号に置換
SpaCyで対象の単語を学習
Singling Out Shingles Vaccine
Singling Out <CONDITION/TREATMENT>
Preprocessing
ステミング
地名・人物名・数値などを記号に置換
Singling Out <CONDITION/TREATMENT> -
13 Health Facts.
single out <CONDITION/TREATMENT> -
<CARDINAL> health facts.
42. 提案手法の流れ2 42
Normalization
Candidate generation
Preprocessing
Candidate generation
seq2seqモデルを使って文章を変換
キーワード (e.g. vaccine) に関連する
クリック率の低い広告と高い広告のペア
から変換方法を学習
single out <CONDITION/TREATMENT> - health fact.
check out <CARDINAL> health fact about
<CONDITION/TREATMENT> on <ORG> right now.
LSTM
seq2seq
LSTM
<CONDITION/TREATMENT> - everything you need
to know. discover <CARDINAL> fact on
<CONDITION/TREATMENT>. get expert advice now!
47. モデル概要 47
推薦対象のユーザ u が like した outfit i と like していない outfit j をサンプリング
①:outfit i に含まれる各衣服を画像を基にベクトルで表現
②:ユーザ u をベクトルで表現
③:outfit i に含まれる各衣服とユーザ u のベクトルを連結して圧縮
④:outfit i に対するユーザ u のスコア (好みの度合い) を計算 (詳細は次スライド)
①
②
③ ④
𝑔𝑔, 𝑓𝑓, 𝜙𝜙:全結合層
50. Track: Crowdsourcing and Human Computation
Becoming The Super Turker: Increasing Wages
Via A Strategy From High Earning Workers
Saiph Savage (Universidad Nacional Autonoma de Mexico (UNAM))
Chun Wei Chiang (West Virginia University)
Susumu Saito (Waseda University)
Carlos Toxtli (West Virginia University)
Jeffrey Bigham (Carnegie Mellon University)
54. Super Turkerのノウハウを使ったnoviceワーカーの支援 54
Experimental groupのワーカーの方が1時間あたりに稼ぐ額が多くなった
約半数のExperimental groupのワーカーは参加基準に全く従わなかった:
ワーカーの技術不足等の問題で参加記述を満たすタスクに参加できず
Control group Experimental group
TurkerView
+
Turkopticon
TurkerView
+
Turkopticon
+
Super Turkerの参加基準
55. Track: Economics, Monetization, and Online Markets
Predicting Drug Demand with Wikipedia Views:
Evidence from Darknet Markets
Sam Miller (University of Warwick & Alan Turing Institute)
Abeer El-Bahrawy (City University London)
Martin Dittus (Oxford Internet Institute, University of Oxford & Alan Turing Institute)
Mark Graham (University of Oxford)
Joss Wright (University of Oxford)
58. Track: Social Network Analysis and Graph Algorithms
Friend or Faux: Graph-Based Early Detection of
Fake Accounts on Social Networks
Adam Breuer (Harvard University)
Roee Eilat (Facebook)
Udi Weinsberg (Facebook)
64. Track: Security, Privacy, and Trust
Detecting Undisclosed Paid Editing in Wikipedia
Nikesh Joshi (Boise State University)
Francesca Spezzano (Boise State University)
Mayson Green (Boise State University)
Elijah Hill (Boise State University)
68. Track: Search
End-to-End Deep Attentive Personalized Item
Retrieval for Online Content-sharing Platforms
Jyun-Yu Jiang (University of California, Los Angeles), Tao Wu (Google)
Georgios Roumpos (Google), Heng-Tze Cheng (Google), Xinyang Yi (Google),
Ed Chi (Google), Harish Ganapathy (Google), Nitin Jindal (Google), Pei Cao (Google),
Wei Wang (University of California, Los Angeles)