eric gilbert | asst prof | georgia tech [Gilbert & Karahalios. - PowerPoint PPT Presentation

eric gilbert | asst prof | georgia tech

[Gilbert & Karahalios. CHI 2009]

Triad Trait Example design problem When chatting with A and C, 1. Visibility A C how does B not highlight the “forbidden triad?” B 2. Visibility B hears something from A A C relevant to C. How does B bring it to C’s attention? B 3. Awareness Are A and C always aware A C that B hears everything they say to one another? B 4. Awareness A can hear what B says to C, A C but not what C says back. * addressed by Twitter. B 5. Awareness A followed C because B did. A C Now B severed the tie. Does A still want to follow? B 6. Accountability B can take credit for what C A C says, since A only hears C through B. B [Gilbert. CHI 2012]

Link Different http://bit.ly/a45Hgb copy who? 3,879 followers 2,190 saw it E A R LI E R http://bit.ly/JT679N 3,879 followers 1,597 saw it Stats from link di fg erent fj rst? build. URL search Text features Twitter search Triad cache bit.ly <title>… fj rst 10,000 text people tinyurl <h*>… people who + is.gd <meta>… reference URLs MySQL goo.gl Google corpus + search text … … … … … machine 1, actors: 512 machine 2, actors: 1,024 machine 3, actors: 1,024 machine 4+5, actors: 1,024 shortened URLs discovered URLs [Gilbert. CHI 2012]

new & ongoing: modeling rumors courtesy noaa

1 Streaming + preprocessing (MC) Time Duration of Collection = 96 days Streaming API : 1% sample D p : Processed Tweets Every 1M tweets Streaming Tweets tracked tweetID tweet_post_time tweet_json via Streaming API D p Tweet lang Remove Tokenize Remove 530884 2014-11-05 19:48:56 {text:..,author:..} 1,086,942,494 code = English Spam Tweets Stopwords . . . . . . . . . (~ 1.09 billion) Processed tokens T s : Topics 2 Online LDA (MC) Total topics annotated topic# word1, word2, word3, ..., word i 46,850 Initialization Iterative Step T s topic 1 p(w 1 |t 1 ) ..... Load . Initialize Run LDA Update . Event/Non Event model Parameters for current topic j p(w 1 |t j ) ..... p(w i-1 |t j ) Parameters annotations t i -1 . α, β, k , c time t i . 468,500 k -Topics E s : Event_Annotations 3 EventAssess (HC) Search 10-Turker Event (if Event) topic# Topics agreed upon Query Responses Assessment Summarization E s Search Boolean AND as Events Send to Turkers for topic 1 q=w 1 &w 2 &w 3 [1,0,-1...1] 1(Event) [ Messi scored, top 3 words Event Assessment . 1049 . Query (per topic) topic j [0,0,0,0..1] 0(NonEvent) . . Search Queries for C s : Credibility_Annotations Agreed upon Events Search 30-Turker Turker Turker topic# 4 CredAssess (HC) Query Responses Reasonings Response Time Total Credibility E s [ News sources topic 1 q=w 1 &w 2 &w 3 [2,2...1] Annotations 2014-11-05 19:48 . confirm,.. . Send to Turkers for . 31,470 topic j [-1,0..2] [ Lots of people Credibility Assessment tweeting,.. S tweets : Searched_Tweets Event speci fj c tweets 5 SearchTwitter (MC) Search S tweets topic# Searched & Collected tweetID tweet_post_time tweet_json Query via Search API topic 1 q=w 1 &w 2 &w 3 530884 {author:..,} Search Tweets 2014-11-05 19:48:56 60,097,152 . . Search API . (~ 60 million) topic j [Mitra & Gilbert. ICWSM 2015]

new & ongoing: censorship on sina weibo https://flic.kr/p/qE2UAj (cc by-sa)

[zheng] ���� � � � ��� ��� ! decompose [fu] ���������� zhèng f ǔ combine & (government) calculate score( �� ) = 149.94 score( �� ) = 139.14 ��� score( �� ) = 136.13 randomize … zhèng f ū (no meaning) [Hiruncharoenvate, Lin, Gilbert. ICWSM 2015]

Q1: How do early design decisions affect future, systemic behavior?

Q2: How do you know whether a social system will work? Achieve critical mass?

[Grevet & Gilbert. CHI 2015]

Q3: Can we get to interpretable models that are understandable by users and designers?

eric gilbert | @eegilbert | eegilbert.org