Sunbelt05

Post on 12-Nov-2014

456 views 0 download

Tags:

description

 

transcript

Social Network Dynamics in the Blogosphere

The Blog Research on Genre (BROG) Project

School of Library and Information Science

Indiana University Bloomington

BROG project members

• Susan Herring• Inna Kouper• Sarah Mercure• John Paolillo• Lois Ann Scheidt• Peter Welsch• Elijah Wright

The Blogosphere

1. The collective term encompassing all weblogs(cf. blog biosphere or ecosystem)

2. The “intellectual cyberspace” inhabited by bloggers(Wm. Quick, 2001)

3. “Blogs as a community; blogs as a social network”(www.samizdata.net)

Previous research

• One-third of blogs have no hyperlinks

• Small part of the blogosphere is densely interlinked

• ‘A-list’ blogs are central in network

• Cliques exist

• ‘Conversation’ between blogs is sporadic over time

(Efimova & de Moor, 2005; Herring et al., 2004, 2005; Kumar et al., 2003)

BUT: No previous research on change over time in blog networks

Research question

• How do networks of links among blogs change over time?– How quickly?– To what extent?– In what ways?

Sampling method

• Random sample of 4 blogs followed by snowball sample out 3 levels from random blogs

• 3 samples at 4-month intervals– April, August, December 2004

– samples 2 and 3 automated

– 5387, 4900, 4367 unique URLS per sample

(~10,000 total unique URLs)

Source blogs

a) pencilinyourhand.blogspot.com

b) www.danm.us/blog

c) www.mysocalledblog.com

d) orangetang.org/erica/blogger.html

Analytical methods

• Content analysis– 300 random, 150 core blogs (17+ in-links)

• Themes: current events, politics, religion, technology, etc.

• Blog type: personal journal, filter, k-log, mixed, other

• Gender of blog author

Results compared for three samples

Analytical methods (cont.)

• Social network analysis (Degenne & Forsé, 1999)

– based on links in sidebars (‘blogrolls’)• Centrality• Reciprocity

• Visualization of network core– blogs with 10+ in-links– Kamada-Kawai layout in R

Results compared for three samples

Content analysis: Random subsample

• Themes– Personal > current events/politics > technology >

religion

• Blog type– Filter - avg. 42%, increasing over time– Personal journal - avg. 38%, decreasing over time

• Gender of blog author– Male - avg. 65%, increasing over time

Cf. Herring et al. (2004)– 70% of blogs are personal journals;13% are filters– 50% of blog authors are female

Content analysis: Core sample

• Themes– Religion > current events/politics > personal >

technology

• Blog type– Filter - avg. 49%, increasing over time– personal journal - avg. 15.6%, decreasing over time

• Gender of blog author– Male - avg. 66%

Core: blogs with 17+ in-links

Content analysis: Comparison

• Random subsample– few in-links (peripheral to network)few in-links (peripheral to network)– diverse contentdiverse content– high turn-over of individual blogshigh turn-over of individual blogs

• 13% shared across 3 samples

• Core sample– many in-linksmany in-links– focused on religion, politics, morality, educationfocused on religion, politics, morality, education– stable membership over timestable membership over time

• 75% shared across 3 samples

Social network analysis: Centrality

• ‘A-list’ blogs are central– All four source blogs lead to 25/37 A-list blogs

– Avg. 3 degrees of separation from any source blog to any A-list blog (range 1.8 - 4.7 degrees)

• tendency to increase in closeness over time

• Catholic blogs are ‘core of the core’– pattern like A-list

Social network analysis: Reciprocity

• A-list blogs attract more links– Tend to be found in reciprocal relations with other A-list blogs

– Non-A-list blogs link preferentially to A-list blogs, but low rate of reciprocation

• Change over time– Increase in reciprocal linking of A-list blogs (p = .001)

– Decrease in reciprocal linking of non-A-list blogs (p = .001)

• Catholic blogs pattern like A-list

Visualization

• Cut-off at 10 in-degrees (350 blogs)• Three thematic clusters emerge:

• Catholicism (red)• Politics/current events (green)• Homeschooling (blue)

• Catholic (and some political) blogs consolidate over time

• Other clusters fragment or disperse

QuickTime™ and aTIFF (LZW) decompressor

are needed to see this picture.

Sample 1 (April 2004)

Sample 2 (August 2004)

Sample 3 (December 2004)

Animation

QuickTime™ and aVideo decompressor

are needed to see this picture.

Study limitations

• Only four random sources, three of them filter blogs, one Catholic

– Filters more likely to have links (Blood, 2002)– Catholic blogs more likely to link to each other?

• Snowball sampling creates bias towards connectivity

– Overestimates overall connectivity

• First sample was collected manually, second and third samples via automated crawl

– May not be strictly comparable

How does the network change?

• Core gets tighter– religious, politically conservative blogs

• Periphery gets looser– thematically-diverse, albeit disproportionately

filter-type, male blogs

• Change is evident at 4-month intervals

Possible explanations

• Political/religious discourses increasingly polarized– US 2004 presidential campaign

• Tendency for cliques to become more cliquish– If so, should be demonstrable for other cliques in

the blogosphere

Future directions

• Conduct longitudinal network analysis starting from other source blogs, e.g.– Politically liberal

– Non-filter types

– Female authors

• Sample at shorter intervals

• Track network evolution over long time spans

Contact:

brog@blogninja.com