LA Adamic, N Glance - 3rd international workshop on Link …, 2005

From ScribbleWiki: Analysis of Social Media

The Political Blogosphere and the 2004 U.S. Election: Divided They Blog

Lada A. Adamic, Natalie Glance

Social media has seen growing relevance in politics. In 2004 itself, it has made is importance evident in a number of ways – like the Howard Deans’s campaign and controversial topics being raked up first and kept alive in the blogosphere.

This paper is amongst first significant step in analyzing the social network of the blogosphere from the political perspective. The authors study the linking patterns between blogs just before the 2004 presidential elections.

Data Collection

The authors first get a list of Liberal and Conservative blog urls from weblog directories. If there are blogs that are consistently linked from these blogs, then those blogs are retrieved too, and manually labeled for its affiliation. For the preliminary analysis, the authors use only the front pages of these blogs

Citation network:

The authors use the blogs thus collected to create a citation network, which is essentially a graph where the blogs are nodes, and the urls (post citation and page links) form the links between them. The citation network indicates a cyber-balkanization tendency. The liberal blogs are clustered together, while the conservative blogs form another close knit cluster. This is because 91% of the time, an emanating link is likely to link to a like-minded blog (right leaning to right leaning, and left leaning to left leaning). Another observation was that the right leaning bloggers were more likely to cite. It was found that 84% of the right-leaning blogs had outgoing links, and 82% incoming links, while for left leaning, only 74% had outgoing and 67% had incoming links.

Analysis of Posts: Data

The authors perform further detailed analysis of the posts in top 20 blogs from each category. The top 20 were selected based on their page links and the citations they got. Posts were collected from these for 2.5 months, which resulted in a dataset of 12470 left leaning, 10414 right leaning posts.

Analysis of Posts: Strength of community

The authors define strength of community based on the number of posts where one blog cited another blog. Here too, on lines similar to citation network, the right-leaning blogs show more connectivity. When the criteria of strength is made more severe (at least 25 links should exist) the right leanign cluster and the left leaning cluster become disconnected components, and the graph of left leaning blogs are very sparsely connected. This draws the conclusion that the right leaning blogs have a denser structure of strong connections than the left

Analysis: Interaction with mainstream media

The authors find that bloggers heavily link to news sites. On an analysis of this links, it was found that right leaning news sites are more often linked to by right leaning news bloggers and left leaning news sites are more often linked to by left leaning bloggers (though this is not a very definite pattern, there are some news sites that have political leanings but are equally cited by both blogger types)

Analysis: response to CBS news item

The authors then analyze the reaction of the bloggers to a controversial news item (that attached George Bush). It was found that the conservative bloggers took up the story and actively blogged about it, while the left leaning bloggers did not blog about it much. This could be perhaps because the conservative bloggers are more vocal, or their different linking patterns cause more information flow, talk and interaction.

Analysis: Occurrences of names of political figures

Analysis of occurrences of political figures in blogosphere showed some very interesting results. Conservatives like Dick Cheney and Colin Powell were more often talked about in the left leaning blogs, while the Liberal figures like Michael Moore and Dan Rather were more often talked of in the conservative blogs. The authors attribute for this by stating that people support their positions by criticizing those of the political figures they dislike. This is an interesting observation, and a similar implicit assumption is used in the Kim and Hovy 2007 paper on election outcome prediction to get automatically labeled data.

In summary, this paper puts forward many interesting observations and analyses of the political blogosphere. There is a clear division of the blogosphere as regards the linking patterns, news items they cite, and people they talk about.