An Analysis of Political Views on Blogs

1
An Analysis of Political Viewson Blogs

• Todd Sullivan

2
General Layout of Project

• 8 Classes
• Candidates Obama, McCain, Biden, Palin
• Parties Republicans, Democrats, Liberals,
  Conservatives
• Extract opinions about each class
• Compute a FeelScore metric for each author/class
  pair (provided the author mentioned the class)
• Do stuff with the FeelScores

3
Extracting Opinions

• Define three lists
• Class Names
• Word Synonyms
• Feeling Indicators
• Tokenize blog posts into sentences
• Find sentence that contain a class name
• Count number of positive/negative Feeling
  Indicators
• Apply counts to classes in sentence

4
Class Names

• Obama
• obama, senator of illinois, senator from
  illinois, illinois senator, democrat president,
  democratic president,
• Palin
• palin, alaskan governor, governor of alaska,
  alaska governor, republican vp,
• Liberals
• liberals, liberal, libs

5
Word Synonyms

• Bad
• worst, unpleasant, disastrous, dirty, failure,
  sucks, sux, traitor, idiot, loser, noob, dumb,
• Good
• awesome, sweet, cool, groovy, comforting,
  glorious, dandy, brilliant, best, wonderful,
• Skip Words
• Very, usually, still, much, nearly, most, more,
  frankly, a, an, any kind of punctuation,

6
Feeling Indicators

• Negative Indicators
• is bad, bad plans, plans is
  bad,i do not want, i cannot stand, how
  bad, is not good, i disagree, i do
  not,
• Positive Indicators
• is experienced, i agree, i will vote,for
  president, i support, i donated,good
  speaker,

7
FeelScore

• For each author a and class c

8
Incorporating Interests and Communities

• Pull all unique interests containing one of the
  class names (obama, mccain, )
• Returns around 350 interests
• Small enough to label as positive, negative, or
  remove.
• Same process for communities

9
(No Transcript)
10
(No Transcript)
11
Interesting Interests

• Positive Interests
• Not interesting obama, mccain 08, republican
• Negative Interests
• _doing_something_ to _class_where
  _doing_something_ isanti, testing, baiting,
  bashing, hating,pissing off, cockpunching,forced
  lobotomization of, death to

12
(No Transcript)
13
(No Transcript)
14
Example Communities

• Positive
• Obama PA, GothsForObama, TeenRepublicans
• Negative
• Anti-Obama, nobama, WTF-Palin

15
Authors Listing Multiple Interests for the Same
Class
16
Authors Joining Multiple Communities for the Same
Class
17
FeelScores Across Time

• For an author a, class c, and day Day

18
Including Interests and Communities

• Count an interest or community as 5 positive or
  negative counts added to AdjCount(a,c,Day,...)
  for all days on and after the interest or
  community was added to the database.

19
FeelScores Across Time Continued

• For a class c, and day Day

20
Overall Blog-basedCandidate FeelScores
21
Overall Blog-basedParty FeelScores
22
Positive/Negative Candidate Chatter by Blog
Network

23
Blog-based Candidate FeelScores Across Networks
24
Blog-based Party FeelScores Across Networks
25
Positive/Negative ChatterCandidate by Gender
26
Blog-based Candidate FeelScores Across Gender
27
Blog-based PartyFeelScores Across Gender
28
National Conventions
29
National ConventionsNew Authors
30
National ConventionsChanging Opinions
31
Presidential Debates
32
Presidential Debates Cont
33
Vice Presidential Debate
34
VP Debate Cont
35
Obama McCainFeelScores by Network
36
National Conventionsby Network
37
McCain PalinFeelScores by Network
38
Party FeelScores by Network
39
Aggregate FeelScoresby Gender
40
Age Ranges
41
Aggregate FeelScores by Age
42
(No Transcript)
43
Older People Dont Like Liberals
44
Friend Networks

• 523 Democrats and 620 Republicans
• 1,575 mutual links
• 64 Democrat-Republican links (Green)
• 23.6 Democrat-Democrat links (Blue)
• 12.4 Republican-Republican links (Red)
• The 64 number is largely influenced by a few
  authors. In a macro-average across authors,
  50.9 of a bloggers friends are from the
  bloggers party.

45
Percent of Bloggersvs. Number of Friends
46
Political BloggingFriend Network
47
National Convention Poll Data
48
Predicting the Popular Vote

• No data after October 28, 2008, so we use Oct. 28
  data for calculations.
• For each author a

49
Assigning Votes Continued
50
Popular Vote

• Actual Result Obama 53, McCain 46

51
Exit Polls by Gender
52
Exit Polls by Age
53
Questions?