Characterizing Silent Users in Social Media Communities Wei Gong, Ee-Peng Lim, Feida Zhu Living Analytics Research Centre School of Information Systems Singapore Management University
Users in social sites can: •
•
Generate content • Create information • Share news • Interact with people Consume content
Silent Users (or Lurkers)
Why do we study lurkers? • •
There are a large number of lurkers online Lurkers have interests and opinions • To avoid conclusion biased by active user • To improve personalized services, recommendation systems, and targeted advertising
Outline •
Define and characterize lurkers
•
Profile lurkers
Outline •
Define and characterize lurkers
•
Profile lurkers
Define Lurkers We say a user is lurking or a user is a lurker during a time interval with duration d , if the number of tweets he/ she posts in the time interval is not more than a lurking threshold h .
Sample Twitter Communities •
Singapore Twitter community size: 110,907 users • All tweets from April 28th to August 31st, 2014
•
Indonesia Twitter community size: 114,576 users • All tweets from June 16th to Oct 19th, 2014
Proportion of Lurkers
(d = 1week)
A significant number of lurkers are found in Singapore and Indonesia Twitter communities
Connectivity
L: lurkers A: Active Users
h=5 d = 18 weeks Singapore Community
Lurkers are less attractive for others to follow, and also are less interested in following others
Why do lurkers break silence?
h=5 d = 18 weeks Singapore Community
When a big event happened, lurkers follow the general trend
Outline •
Define and characterize lurkers
•
Profile lurkers
Profile Lurkers •
Predict lurkers’ attributes (marital status, religion, and political orientation) using content features: • The user’s tweets • The user’s followees’ tweets • The user’s followers’ tweets • The users’ mutual friends’ tweets
•
Datasets
Performance of Profiling Martial Status
All Users
Lurker
Active Users
Profiling lurkers can achieve • comparable performance to that of profiling all users • better performance than that of profiling highly active users
Performance of Profiling Religion and Political Orientation Religion
All Users
Lurker
Political Orientation
Active Users
All Users
Lurker
Active Users
Profiling lurkers can achieve comparable performance to that of profiling active users
Summary •
We underscore the presence of lurkers
•
We characterize and profile lurkers in Twitter communities. We show that profiling lurkers can be as accurate as profiling active users
•
In the future, we could uncover lurkers’ interests and opinions.
Email:
[email protected]
Supporting Slides
Twitter Community Sampling Method •
Start the crawling process from a set of seed users (from Singapore or from Indonesia), and add users who are one hop and two hops away from the seed users •
•
140,850 Singapore users and 126,047 Indonesia Users
Make sure the users we study do not churn: • Remove the users who do not generate any content for the following 3 months •
110,907 Singapore users and 114,576 Indonesia Users