The Geospatial Characteristics of a Social Movement Communication Network
1. The Geospatial Characteristics of
a Social Movement Communication
Network
By Karissa McKelvey
Joint work with:
MD Conover, C Davis, E Ferrara,
F Menczer and A Flammini
8. How does information about protest
spread across geographically
decentralized online
communication networks?
9.
10.
11. Hypotheses
• H1: Information about the protest is
disseminated from prominent protest areas
• H2: Communication between locals exhibit
different signals than long-distance
communication
17. Data Collection
Occupy
#ows and #occupy*
1.5M tweets
>250k users
Domestic Politics
#tcot and #p2
825K tweets
>68k users
July 3rd, 2011 to March 12th, 2012
21. Geolocation Detection
• State-level data
– “New York, NY” -> “NY”
– “Earth” -> “N/A”
• Detect locations of 55.7% Occupy and 29.3%
Domestic politics users
• 53% are associated with a location among the
1,000 most popular self-reported responses
22. Geolocation Detection Method
• Black List
– Hand-coded 1,000 popular strings
– e.g. “the dance floor” “Earth”
– 161 locations, or 6% of the tweets
• String “fuzzy matching”
– Misspellings
– Similarities, e.g. “on the dance floor”
23. Location Results
Occupy Politics
Total Users 257,000 68,000
Self-reported 174,760 (68%) 24,480 (36%)
Location
Detected
97,341 (38%) 7,099 (10%)
U.S. State 36,794 (22%) 5,849 (8%)
24. How does the information travel,
and from where?
25.
26.
27. Proportion of Retweet Traffic by State
State (Ordered by Maximum for Each State)
ProportionofTotalTraffic
0.00
0.05
0.10
0.15
0.20
0.25
NewYork
California
DistrictofColumbia
Florida
Texas
Illinois
Kentucky
Massachusetts
Wisconsin
Oregon
NewJersey
Alabama
Georgia
Pennsylvania
Michigan
Arizona
Virginia
Washington
Colorado
NorthCarolina
Minnesota
Missouri
Maryland
Nevada
Ohio
Oklahoma
SouthCarolina
Indiana
Tennessee
Montana
Louisiana
NewHampshire
RhodeIsland
Kansas
Hawaii
Alaska
Utah
Mississippi
Wyoming
Iowa
Maine
NewMexico
Connecticut
Arkansas
Vermont
Idaho
Delaware
Nebraska
WestVirginia
NorthDakota
SouthDakota
stream
Domestic
Occupy
32. Summary
• Occupy discourse on Twitter has highly localized
geospatial structure: a large amount of traffic is
produced and consumed locally per state.
• Interstate communication is driven by high-profile
locations acting as information broadcasters.
• Proximity to on-the-ground events plays a big
role: users from NY, DC and California are the
main actors of the discourse.
33. Further Thoughts
• Methodological contributions
• Future work could investigate if other social
movements exhibit different behaviors.
• Twitter could very well be an important tool
for catalyzing social movements, but more
work needs to be done
34. Thanks!Papers at cnets.indiana.edu/groups/nan/truthy
Sandro Flammini
Bruno Conçalves
Jacob Ratkiewicz
LilianWeng
Mike Conover
Johan Bollen
KarissaMcKelvey
Przemek Grabowicz
Mark Meiss
AlexVespignani
Alex Rudnick
LucaAiello
Fil Menczer
Mohsen Jafari-Asbagh
Onur Varol
Emilio Ferrara
Wednesday, September 26, 12
When protests were organized in the past, during the so-called civil rights era, or even far before this and in other countries,
We saw individuals and institutionsorganizing by way of fax; recruiting new supporters by way of mail, and disseminating their message through newspapers and press releases.
Social movements are complex, political networks. The world has been captivated by hundreds of thousands of people entering the streets in protest during the arab spring as well as recent protests in turkey and brazil.
In the united states, the grass-roots Occupy movement sprang up in thousands of cities around the world. These protests seem to have sprang up spontaneously, quickly, with a unified message – 99%, “banks got bailed out, we got sold out”, and other slogans were common. But How did these protests organize?
Recent literature, and the intuitions of many protestors and observers, suggest that online social media have acted as catalysts for these protests
Social media can fosters the creation of international networks between millions of amateur “citizen journalists” activists and others.As we use these technologies, we leave digital traces that can be collected on a massive scale.We can collect the communication, organization, and information dissemination mechanisms, that movements use in a very public setting. For example, on Foursquare, users checked in to occupy encampments and encouraged others to participate in the protests, disseminating the location, time, and place of future events (for example).
How do we measure this spread of protest?
There are two separate social processes that have been identified in social movement communications.Collective framing: the social processes whereby movement participants negotiate the shared language and narrative frames that help define the movement's identity and goals.Resource mobilization: the work to marshal the physical and technological infrastructure, human resources, and financial capital necessary to sustain ongoing activity.Broader ideas of mechanisms within social movements
Motivations of the protestSocial and wealth inequalities, taxation disparitiesNon sustainable capitalistic market modelsPolitical corruption, corporate influence of government
We are constantly collecting an 8% random sample of Twitter data, and have been since July, 2010.
Retweeting is a mechanism of endorsement. When users retweet another, they send the information of that tweet to their followers’ social feeds. This is different than the mechanism of mentioning, which is when users are talking to, or about eachother on the platform. In this analysis, we are primarily focusing on the retweet mechanism. Future work might do well to study the in-group and out-group conversation.
We limit our study to the United States. We first want to compute where people are tweeting from, and extract their retweet networks.Each node is a state, and each edge is a retweet from one user to another in different, or the same, states. This includes cycles, so
We could just study the social movement’s communication alone, but we wanted to get a better picture as to how this communication was different than normal Twitter traffic.SO, we decided to analyze Occupy communication networks in comparison to those of Domestic Politics.
Over 500 million users and growing wordlwideLatitutide, longitude
geo
Thousand most popular strings manually, blacklisting those that didnot correspond to geographically meaningful entities. Drawn from a long tailed distribution, 53% ofall tweets in the data set are associated with a location among the 1,000 most popular responses, with27% of all tweets containing one of the top hundred location strings. From this set of one thousand weblacklisted 161 non-location strings, corresponding to 6% of the tweets associated with the 1,000 mostpopular responses.
NY, California and DC are producing most of the traffic proportionally to other states.Some states very active in political discourse, such as Kentucky or Alabama, show little to no interest in Occupy-related topics.
The intensity in the color represents how much the amount of Occupy-related traffic deviates from that of domestic politics per state.Maine and Oregon, outliers
Occupy-related discourse (on the right) shows a prominent hub-and-spoke structure differently from domestic politics (on the left).Multiscale backbone extraction – confidence level a = 0.15
Occupy communication patterns exhibit heightened local activity. More than 3 times than discussion about U.S. politics.
They produce much more Occupy-related information than that they consume, unlike other states.