Visual Analysis of Topic Competition on Social Media
Post on 10-Nov-2014
1952 Views
Preview:
DESCRIPTION
Transcript
1
Panpan Xu1, Yingcai Wu2, Enxun Wei2, Tai-Quan Peng3, Shixia Liu2, Jonathan J.H. Zhu4, Huamin Qu1
……
…Visual Analysis of Topic Competition
1 Hong Kong University of Science and Technology
2 Microsoft Research Asia
3 Nanyang Technological University
4 City University of Hong Kong
on Social Media
VAST 13
2
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Diffusion of multiple topics
The Interaction: Do people get distracted away from some topics when something more “eye-catching” is happening?The Influence: How do the opinion leaders (influential users) affect the interaction by recruiting the public attention for some topics?
On Social Media:
Google Ripples
3
Whisper [N. Cao et al. 12]
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Google Ripples [F. Viégas et al. 11]
4
Agenda-setting The ability of the news media (e.g. TV and newspaper) to influence the salience of topics on the public agenda.
Topic competition
Two-step information flow
The addition of any new topic onto the public agenda comes at the cost of other topic(s).
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
[M. E. McCombs and D. L. Shaw 72]
[J. Zhu 92]
[S. Wu et.al 11]
The information reaches the masses via intermediaries.
5
Agenda-setting The ability of the news media (e.g. TV and newspaper) to influence the salience of topics on the public agenda.
Topic competition
Two-step information flow
The addition of any new topic onto the public agenda comes at the cost of other topic(s).
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
[M. E. McCombs and D. L. Shaw 72]
[J. Zhu 92]
[S. Wu et.al 11]
The information reaches the masses via intermediaries.
6
Agenda-setting The ability of the news media (e.g. TV and newspaper) to influence the salience of topics on the public agenda.
Topic competition
Two-step information flow
The addition of any new topic onto the public agenda comes at the cost of other topic(s).
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
[M. E. McCombs and D. L. Shaw 72]
[J. Zhu 92]
[S. Wu et al. 11]
The information reaches the masses via intermediaries (opinion leaders).
7
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Combine quantitative modeling and interactive visualization
Healthcare
#debate
#chinaExtract time varying measurements on • topic competitiveness• each opinion leader group’s influence on each topic• topic transition trend of each opinion leader group
Visualize • the dynamic relation between topics and opinion leader groups• textual contents of the posts
8
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Collection of Tweets
Text Search
Time Series: Stream of
tweets
TopicUser group
Topic Competition
Modeling
Topic Transition Analysis
Combine quantitative modeling and interactive visualization
TimelineVisualizatio
nRaw Tweets
List
Word Cloud
9
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Collection of Tweets
Text Search
Time Series: Stream of
tweets
TopicUser group
Topic Competition
Modeling
Topic Transition Analysis
Combine quantitative modeling and interactive visualization
TimelineVisualizatio
nRaw Tweets
List
Word Cloud
10
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Collection of Tweets
Text Search
Time Series: Stream of
tweets
TopicUser group
Topic Competition
Modeling
Topic Transition Analysis
Combine quantitative modeling and interactive visualization
TimelineVisualizatio
nRaw Tweets
List
Word Cloud
• Time-varying topic competitiveness
• Each opinion leader group’s influence
• Topic transition trend
11
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Collection of Tweets
Text Search
Time Series: Stream of
tweets
TopicUser group
Topic Competition
Modeling
Topic Transition Analysis
Combine quantitative modeling and interactive visualization
TimelineVisualizatio
nRaw Tweets
List
Word Cloud
12
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
∆ 𝑝𝑖𝑡=𝑚𝑖
𝑡−1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽𝑖𝑗𝑝 𝑗𝑡− 1−𝑝𝑖
𝑡 −1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽 𝑗𝑖𝑚 𝑗𝑡 −1
[J. Zhu 92]
distraction effect
Topic Competition Model for traditional media:
recruiting effect
change of public attention on topic
13
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
∆ 𝑝𝑖𝑡=𝑚𝑖
𝑡−1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽𝑖𝑗𝑝 𝑗𝑡− 1−𝑝𝑖
𝑡 −1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽 𝑗𝑖𝑚 𝑗𝑡 −1
recruiting effect
[J. Zhu 92]
media coverage on topic
population on other topic
Topic Competition Model for traditional media:
change of public attention on topic
14
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
∆ 𝑝𝑖𝑡=𝑚𝑖
𝑡−1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽𝑖𝑗𝑝 𝑗𝑡− 1−𝑝𝑖
𝑡 −1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽 𝑗𝑖𝑚 𝑗𝑡 −1
[J. Zhu 92]
distraction effect
population on topic
media coverage on other topic
Topic Competition Model for traditional media:
change of public attention on topic
15
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
∆ 𝑝𝑖𝑡=𝑚𝑖
𝑡−1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽𝑖𝑗𝑝 𝑗𝑡− 1−𝑝𝑖
𝑡 −1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
𝛽 𝑗𝑖𝑚 𝑗𝑡 −1
∑𝑔=1
𝑛
𝑚𝑖 ,𝑔𝑡− 1
The Extended Topic Competition Model:Two step information flowHeterogeneous influence (news media, grassroots)
16
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
𝑝𝑖𝑡=𝑎𝑖𝑝𝑖
𝑡−1+∑𝑔=1
𝑛
𝑚𝑖 ,𝑔𝑡− 1 ∑
𝑗=1 , 𝑗≠𝑖
𝑘
𝛽𝑖 , 𝑗 ,𝑔𝑝 𝑗𝑡 −1−𝑝𝑖
𝑡−1 ∑𝑗=1 , 𝑗≠ 𝑖
𝑘
∑𝑔=1
𝑛
𝛽 𝑗 ,𝑖 ,𝑔𝑚 𝑗 ,𝑔𝑡− 1
The Extended Topic Competition Model:Two step information flowHeterogeneous influence (news media, grassroots)
distraction effectrecruiting effect
Topic competiveness& opinion leader’s influencethrough decomposition
17
min∑𝑙
𝜔𝑙‖𝑚𝑙𝑡 −1𝐴−𝑚𝑙
𝑡‖2
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜:∑𝑗=1
𝑘
𝑎𝑖𝑗=1𝑎𝑛𝑑𝑎𝑖𝑗≥0
topic
topic
topic
topic
topic
topic
topic
topic
…
…
…
…
Topic Transition Estimation
𝐴𝑘×𝑘=(𝑎11 … 𝑎1𝑘… … …𝑎𝑘1 … 𝑎𝑘𝑘
)Transition matrix
TT-1
18
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Collection of Tweets
Text Search
Time Series: volume of
tweets
TopicUser group
Topic Competition
Modeling
Topic Transition Analysis
TimelineVisualizatio
nRaw Tweets
List
Word CloudOutput of Analysis and Modeling Step:
Time varying competitiveness of each topicTime varying opinion leader groups’ influence on each topicThe topic transition trend of the opinion leader groups between adjacent time stamps.
19
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Topic competiveness
Timeline view
20
Media
Political Figures
Grassroots
Topic competiveness
+ Recruitment effect of different opinion leaders
+ Topic transition trend
Timeline view
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Word Cloud
21
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Word cloud filterable by:• Topic• Time interval• Opinion leader group
Sparkline:• Time varying saliency of a word
22
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Dataset:
2012 Presidential Election; 89, 174, 308 tweets; May 01 – Nov 10Six general topics : welfare/society, defense/international issues, economy, election (general), election (horse race), law/social relations *Three opinion leader groups: media , political figures, and grassroots *
*identified collaboratively with media researchers
23
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
Election (general)
24
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
25
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
26
INTRO / SYSTEM / MODEL / DESIGN / CASE STUDY
27SUMMARY / LIMITATIONS & FUTURE WORK
Collection of Tweets
Text Search
Time Series: volume of
tweets
TopicUser group
Topic Competition
Modeling
Topic Transition Analysis
TimelineVisualizatio
nRaw Tweets
List
Word Cloud
Visual analysis framework:
Model the topic competition on social media, the influence of opinion leader groups, and the topic transition trends.
Visualize the results of the models and allow for further exploration to form explanations.
28SUMMARY / LIMITATIONS & FUTURE WORK
Manual process to collect keywords and categorize opinion leadersmore efficient ways?
Time series modeling+ the structural factors of social network ?
Competition & cooperationother modes of interaction among topics?
29
Thank You for Attention !
top related