Modeling Malaysian Public Opinion by Monitoring the Malaysian Blogosphere

1
Modeling Malaysian Public Opinion by Monitoring
the Malaysian Blogosphere

• Brian Ulicny
• VIStology, Inc.
• Framingham, MA

2
IBlogs Goals

• Cluster Posts by Event, and Rank using new
  metrics for
• Relevance
• Timeliness ( Locality)
• Specificity
• Credibility
• Applied to Malaysian Blogosphere as a Case Study
• AFOSR funded Distributed Intelligence Program
• Automated Ontologically-Based Link Analysis of
  International Web Logs for the Timely Discovery
  of Relevant and Credible Information
• Ends November, 2008.

3
Commercial State of Art

• Technorati Blog Search
• Organized by Date and Authority.
• Authority Global Link Popularity.
• No way to filter blogs by location currently.
• So must probe via keywords.
• No aggregation of topics or links locally.

4
The Malaysian Blogosphere

• Former British colony.
• English 2nd Most Common Language.
• Complex ethnic and religious social structure.
• Press tightly controlled
• RSF 124th of 169.
• But Internet growth encouraged.
• Over 150K Malaysian profiles on Blogger.com.
• 310K profiles on Facebook.com
• Blogs have exposed alleged government corruption
  and organized protests.
• Government has intimidated and arrested bloggers.

5
Malaysian SoPo Blogosphere
6
Malaysia SoPo Blogs385 blogs in crawl seed
yields 220.3K urls, 4693 Sites _at_ depth 4

• 2K Blogs
• 1,059 Blogspot Blogs, 801 Wordpress, 16 Typepad
• 272 .org, 45 .gov.my
• Of the Blogspot blogs (structured profiles)
• 659 Malaysia, 166 other, 234 unspecified
• 122 Kuala Lumpur
• 88 Selangor (State)
• 31 Johor (State), 10 Subang Jaya, 11 Klang, 6
  Johor Bahr
• 10 list Koran as favorite book 4 Bible

7
Fragmentation of ReferencesDiversity of Links
for Bersih Rally Posts
8
Posting and Linking PatternsMalaysian SoPo
Bloggers, Blogging Unfree Press
Cf. 50 of US A-list Blog posts contain a news
link (Adamic, Glance. 2005)
9
Clustering by EventNov. 10, Kuala Lumpur,
Malaysia, police
BBC
Star
CNN
NST
10
Blog Post RelevanceAboutness Determined by Term
Frequency in Post Text plus Linked Text
? Links To
Jeff Ooi Post
IHT article
11
Blog Post SpecificitySpecificity Determined by
Unique NPsin Post Text and Linked Text
Jeff Ooi Pos(26 UNPs)
IHT article (16 UNPs)
? Links To
12
Blog Post Timeliness LocalityTimeliness is
Difference between Post Time and Dateline
13
Centrality vs. Credibility
14
Conclusions and Future Work

• 1. It is possible to geolocate many Malaysian
  bloggers explicitly, particularly Blogspot
  bloggers.
• 2. System can now extract (nearly all) the
  features we need automatically now.
• 3. Integration efforts ongoing to tie all the
  pieces together.
• 4. This will allow us to do a complete, fully
  automated analysis of important events in the
  Malaysian (sopo)blogosphere, such as the Bersih
  Rally example that was repeatedly used in the
  slides.

15
Questions?
Brian Ulicny VIStology, Inc. bulicny_at_vistology.com
16
IBlogs Architecture
Event ontology
Blog Ontology
17
BaseVISor and Ontology Reasoning
Alias Pak Lah?
Abdullah Badawi Pak Lah Abdullah Ahmad Badawi
18
Bersih Rally in Global Blogosphere1432 Posts in
Week 1
19
Bersih Rally Posts
Bersih Rally
Hindraf Rally