O artigo abaixo mostra, por meio do rastreamento de 600 mil mensagens do Tweeter, feitas por participantes do movimento, que tipo de mensagem passavam, com que propósito e para quem.
Os autores descobriram que as características da comunicação refletiam, nas comunicações individuais, os esforços de mobilização em nível local (convocar os atos, divulgar os locais de concentração e as orientações práticas de ocupação) e, na comunicação não individualizada, a busca por desenvolver quadros narrativos do propósito coletivo do movimento em nível nacional.
The organizing forces underlying successful social movements have been studied extensively by sociologists and political scientists [3]–[7]. From this body of work common themes have emerged, include the problems of resource mobilization and collective framing, which together constitute two of the core issues any social movement must address in order to effect social or political change. Resource mobilization refers to the process by which a social movement must marshal the financial, material, and human resources required to sustain its activities [8]. Collective framing is a process whereby the constituents of a social movement, through formal or informal processes, come to establish the narratives, language, and imagery that capture the essential features of the movement’s purpose and struggle [9]. Effective framing helps to foster a sense of community and engagement, and can be a powerful response to countervailing social pressures from establishment organizations [10].
Here we study Occupy Wall Street, a social movement focused on issues relating to the uneven distribution of wealth, social inequality, corporate greed, and the regulation of major financial institutions. Since the first protest on September 17
,
2011, a major feature of the movement has been the long-term physical
occupation of high-visibility encampments, often found in parks, banks,
libraries and foreclosed homes. As a result, the Occupy movement
requires substantial supporting infrastructure, including housing and
sanitation facilities, as well as access to communication technologies.
In spite of this, Occupy has sustained a lasting presence in American
cities including New York City, Oakland, Washington, D.C., and Boston,
which also represent key loci of decision making and protest activity [1], [2].
Under the Occupy model, proposals are brought to a vote before a
general assembly, a form of direct democracy in which any participant is
free to comment or vote on any proposal under consideration. The most
prominent among these organizational structures is the New York City
General Assembly, which has been responsible for producing policy and
key narrative frames such as the popular protest slogan, “We are the
99%,” which references the disproportionate concentration of wealth
among the top 1% of the world’s population [11].
Social media have played a prominent role in facilitating communication and coordination throughout the development of the Occupy Wall Street movement. For example, the first call to action in the Canadian anticapitalist magazine ‘AdBusters’ used the Twitter ‘hashtag’ #occupywallstreet as one of just ten words featured in a full-page ad. Ever since, the Twitter platform has been used extensively by movement participants [2], with #ows being one of the hundred most popular hashtags on Twitter for the year 2011. In addition to Occupy, Twitter has also played a prominent role in several foreign social movements, most notably in the Egyptian revolutionary protests of 2011 [12]–[14].
In this work, we seek to understand the relationship between the geospatial dimensions of social movement communication networks and the organizational pressures facing such movements. To accomplish this, we use a state-of-the-art location inference technique to model relationships among users as a weighted directed network of communication flows between states, in which the weight of each edge corresponds to the volume of traffic between pairs of locations. Using this framework we investigate three distinct relationships: attention allocation and proximity to on-the-ground events, resource mobilization and localized information sharing, and the role of collective framing in long-distance communication.
With respect to the issue of attention allocation, we find that compared to stable domestic political communication the Occupy Wall Street movement exhibits very high levels of geographic concentration, with users in New York, California, and Washington D.C. producing more than half of all retweeted content. Aside from these hubs, however, we find that the appeal of content relating to Occupy Wall Street has a disproportionately local audience. With extended, high profile encampments and large-scale protest action playing central roles in the Occupy movement, we propose that this structural feature reflects the importance of mobilizing human resources at the local level.
Finally, we report on evidence indicating that the content of communication at the national level is distinct from the content of communication among users in the same state. Comparing intrastate versus interstate communication, we find that the terms most overrepresented in interstate communication relate to the movement’s core framing language and the news media, while the terms most overrepresented in local communication reference physical places, protest action, and specific times. These results support the hypothesis that local-level communication activity is driven by the challenge of resource mobilization, while long-distance communication is more strongly associated with collective framing processes.
-character messages containing text and hyperlinks, called tweets,
and interact with one another in a variety of ways. Communication on
Twitter is characterized by directed, non-reciprocal social links that
allow users to subscribe to the stream of content produced by another
user. The content produced by every user an individual follows is
aggregated into a single streaming feed, from which an individual can
selectively rebroadcast content to his or her followers by choosing to
retweet it. In this way, a retweet serves to broaden the potential
audience of a piece of content, and signifies that information has been
transmitted between two individuals. Hashtags, short tokens prepended
with a pound sign (e.g. #taxes or #obama), constitute another important
feature of the platform, and allow the content produced by many
individuals to be aggregated into a custom, topic-specific stream
including all tweets containing a given token.
, 2011 and March 
,
2012 – a nine month period including the birth and maturation of the
Occupy Wall Street movement. The gardenhose provides an approximately 
sample of the entire Twitter stream in a machine-readable format.
Gardenhose tweets include useful metadata, among them a unique tweet
identifier, the content of the tweet (including hashtags and
hyperlinks), a timestamp, the username of the account that produced the
tweet, a free text ‘location’ string associated with the originating
user’s profile, and for retweets, the account names of the other users
associated with the tweet. Tweets from geolocation-enabled mobile
devices also report latitude/longitude coordinates, however the
incidence rate of tweets with this data is not enough to be useful as a
feature in general.
To isolate a representative sample of Occupy Wall Street content we flagged for collection any tweet containing hashtags associated with the Occupy movement, including #ows and #occupy{*} (e.g. #occupywallst, #occupyboston, etc.). To provide a baseline against which to compare our observations, we also extracted content originating from the two most popular communication channels associated with stable domestic political communication, #tcot (Top Conservatives on Twitter) and #p2 (Progressives 2.0). In total, this sampling procedure produced 1,522,415 tweets associated with Occupy Wall Street and 825,262 tweets associated with domestic political communication. As this analysis is concerned primarily with information spreading processes we consider only retweet events from this corpus, resulting in 676,369 retweets among 257,657 users associated with Occupy Wall Street, and 259,703 retweets among 68,049 users associated with stable domestic political communication. Henceforth, we consider these corpora to constitute representative samples of retweet interactions among users participating in the streams of content associated with the Occupy Wall Street movement and stable domestic political communication in the United States.
A caveat to this technique, however, is that it relies on raw text generated by a broad swath of the Twitter population, and so we find geographically meaningless location descriptors included in the dataset. To address this issue we rely on an extensive hand-curated blacklist of popular non-geographical responses such as ‘everywhere’ and ‘the dance floor’. To produce this list we sorted all location strings by popularity and reviewed the thousand most popular strings manually, blacklisting those that did not correspond to geographically meaningful entities. Drawn from a long tailed distribution, 53% of all tweets in the data set are associated with a location among the 1,000 most popular responses, with 27% of all tweets containing one of the top hundred location strings. From this set of one thousand we blacklisted 161 non-location strings, corresponding to 6% of the tweets associated with the 1,000 most popular responses.
To improve recall in the presence of novel input, we used a modified version of the Ratcliff-Obershelp algorithm [35] to detect fuzzy matches between free text location strings and the blacklist of popular non-location responses. As a result, because ‘the dance floor’ is in the set of blacklist responses, strings taking a slightly modified form, such as ‘on the dance floor,’ will also be classified as invalid input. The hand-coded blacklist combined with the Ratcliff-Obershelp fuzzy matching technique resulted in 9% of the free-text location strings being classified as non-location input.
From among the remaining responses we submitted location strings to the Bing.com geocoding API, which returns a best-guess estimate for the corresponding physical coordinates. This output is hierarchically formatted to describe the finest level of geographic resolution available. For example, if a user reports ‘Logan Square, Chicago’ as his or her location, the Bing API will return information about the likely zip code, city, state and country associated with that location. However, if the user reports only ‘USA,’ the information provided by the API describes only a country-level guess as to the user’s location. Owing to decreased coverage at the city-level and the proportionately few users associated with each individual city, we utilize the state-level location estimates for the geospatial components of this analysis.
In total, 68.4% of Occupy Wall Street users reported location strings, and from these we were able to obtain geolocation estimates for 55.7% of these accounts. Among this set of users, 60% of the resulting geolocation estimates included state-level metadata. Response rates were somewhat diminished for users associated with the stream of domestic political communication, with 36% of individuals reporting free-text location strings. Using the procedure described above, we were able to obtain geolocation estimates for 29.3% of all users in the domestic political communication stream, 82.4% of which contained state-level metadata.
, where 
is the total number of retweets originating from state 
and 
is the total number of retweets originating from all states. As
outlined above, we focus on retweets as they correspond to attention
allocation rather than total content production volume.
In addition to the distribution of activity across individual states we examine the information sharing relationships among users in different locations. To accomplish this, we rely on a network representation to characterize the flow of information on Twitter. Taking users as nodes, we define a weighted directed network in which an edge with weight
is drawn from node 
to 
in the event that user 
retweets user 
times. The intuition underlying this approach is that each retweet
provides evidence suggesting that information produced by user 
was evaluated and acted upon by user 
.
Combining the user-level geocode metadata previously described with the network representation defined here we can induce another network describing the volume of communication between users in each state. In this network, nodes represent states, and weighted directed edges are drawn among them. The weight of the edge from
to 
is defined as the sum of the weights among all edges originating from users in state 
and terminating in state 
.
We note, however, that this induced network must have geolocation
labels for each node in a dyad. In the Occupy Wall Street stream we
identify 143,437 tweets for which both the source and target have
state-level geolocation data and 78,467 likewise restricted tweets in
the stream of stable domestic political communication.
, in a tweet from a given class, 
, as 
. Comparing these probabilities yields a ratio, 
,
a value which is large when a token is more common in intrastate
traffic than interstate traffic and small under the opposite conditions.
Figure 1. Proportion of retweet traffic associated with each state, for each content stream.
Ordered by the amount of traffic associated with each state, it is clear that a few high-profile locations serve as the dominant sources of content in the Occupy stream. This concentration stands in contrast to the more heterogeneous activity profile for the stream of domestic political communication.
doi:10.1371/journal.pone.0055957.g001
Figure 2. Divergences in geographic distribution of users.
This cartogram uses color to represent the extent to which the number of Occupy Wall Street tweets in each state deviates from the domestic political communication baseline, computed as:
.
Redder colors indicate that proportionally more Occupy content
originated from the associated state, while whiter colors indicate the
opposite. To minimize the effect of outliers on the visualization and to
highlight variation between states, colors for Maine and Oregon have
been fixed, indicating that the deviation from baseline is more than
three times the expected rate.
doi:10.1371/journal.pone.0055957.g002
We also study the ratio of content production to content consumption by locale. Figure 3
shows this ratio, defined as the total number of retweets originating
from users in that state divided by the total number of tweets retweeted
by users in that state. This value serves to highlight the extent to
which users in a given location are functioning as content producers or
content consumers. Inspecting this plot, we find that in the Occupy
stream users from just five states produced more content than they
consumed. This stands in contrast to the stream of stable domestic
political communication, in which fourteen states exhibit a ratio
greater than one.
Figure 3. Ratio of content production versus content consumption, by stream.
Occupy Wall Street users, by state, exhibit a lower content production to consumption ratio relative to users in the domestic political communication stream. The disproportionately high ratio observed for Kentucky can be attributed to the activity of a prolific, highly popular left-leaning user from that state.
doi:10.1371/journal.pone.0055957.g003
To
highlight the effect of this geospatial concentration on communication
flows between states it is instructive to visualize the structure of
these networks. However, owing to the geographic aggregation process
outlined in section Geographic Profile both networks are highly
dense, with edges spanning most pairs of states. To address this issue
we utilize a technique known as multiscale backbone extraction [36],
which is useful for identifying statistically significant edges in
weighted networks, regardless of the absolute value associated with the
weight of that edge. This technique selects for edges with weights
significantly above the expectation given by an analytically defined
probability distribution that models a random allocation of each node’s
strength among its adjacent edges. Parameterized by a confidence level
factor, 
,
this technique allows for the selection of statistically significant
edges across all weight scales, a feature that is especially valuable
when working with networks with heterogeneous weight distributions such
as those associated with communication or human mobility.
Applying this technique to both networks reveals a communication backbone for the Occupy network that exhibits the highly concentrated hub and spoke structure described above. Figure 4 shows that the Occupy Wall Street network is characterized by minimal state-to-state connectivity, with the majority of statistically significant traffic flowing to and from New York, California and Washington D.C. This is in contrast to the communication backbone for the network of domestic political communication, in which we observe extensive interactions among many pairs of states.
Figure 4. Multiscale backbone (
) of the continental interstate communication networks.
Stable domestic political communication is shown at left, Occupy Wall Street at right. Edges adjacent to New York, California, and Washington D.C. are shown in red. Note that Occupy Wall Street exhibits a clear hub-and-spoke pattern, with the majority of traffic flowing to and from New York, California, and Washington, D.C. Likewise, observe that the Occupy Wall Street network exhibits diminished levels of interstate connectivity compared to the network of domestic political communication. We note that the structure of this network backbone is robust to different parameterizations of alpha.
doi:10.1371/journal.pone.0055957.g004
Figure 5. Connectivity matrices describing directed interstate communication volume.
The edge weight corresponding to each cell is mapped to a color hue on a logarithmic scale ranging from white for edges with the least weight to black for edges with the most weight. The strong diaonalization and limited off-diagonal mass apparent in the Occupy Wall Street matrix is indicative of highly localized communication activity.
doi:10.1371/journal.pone.0055957.g005
Table 1. Lists of tokens most overrepresented in intrastate and interstate communication.
doi:10.1371/journal.pone.0055957.t001
With respect to the finding that the majority of widely rebroadcast content is produced by users in a small number of high profile locations, we observe that these states represent sites of major encampment and decision making activity. Despite the fact that all users can contribute equally to the Occupy stream, it appears that proximity to events on the ground plays a major role in determining which content receives the most attention. This is in contrast to the stream of domestic political communication, in which content from users across the United States is allocated a significant share of attention. Where the stream of domestic political communication looks more like a conversation taking place at the national level, the structure of the Occupy stream is more akin to a broadcast, with just a few locations playing the role of net content producers.
Finally, we propose that interstate communication plays a significant role in the propagation of narrative imagery associated with collective framing processes, and that intrastate communication is driven more predominantly by the pressures of resource mobilization. Looking to the lists of tokens most overrepresented in each type of traffic (Table 1), we find that those more common in interstate communication include references to core framing language and the news media. This finding suggests that when users engage in communication across state boundaries they allocate proportionately higher levels of attention to speech associated with collective framing processes. In contrast, the tokens more common in intrastate traffic relate to protest action and specific times and places. From this we conclude that the content of intrastate tweets deals much more frequently with rallying the movement’s participants, a core function of resource mobilization.
The findings outlined in this paper dovetail nicely with established literature on social movement theory. However, statistical measures are limited in the extent to which they can accurately represent nuanced features of communication, and future work in this domain would benefit from rigorous qualitative content analysis. Moreover, there remains room to improve our understanding of how closely the structure of social media communication mirrors that of other forms of communication. For example, Mislove, et al. found that the geographical distribution of Twitter users tends to over-represent populous counties and metropolitan areas, suggesting that entire rural regions may be significantly under-represented – with similar findings holding true for ethnicity and gender as well [37]. In this respect as well, work of this nature would benefit from deeper involvement from scholars in the social sciences, and we hope that this type of interdisciplinary collaboration will become increasingly common.
Para seguir o blog e receber postagens atualizadas, use a opção "seguir", ao lado.
Os autores descobriram que as características da comunicação refletiam, nas comunicações individuais, os esforços de mobilização em nível local (convocar os atos, divulgar os locais de concentração e as orientações práticas de ocupação) e, na comunicação não individualizada, a busca por desenvolver quadros narrativos do propósito coletivo do movimento em nível nacional.
Abstract
Social movements rely in large measure on networked communication technologies to organize and disseminate information relating to the movements’ objectives. In this work we seek to understand how the goals and needs of a protest movement are reflected in the geographic patterns of its communication network, and how these patterns differ from those of stable political communication. To this end, we examine an online communication network reconstructed from over 600,000 tweets from a thirty-six week period covering the birth and maturation of the American anticapitalist movement, Occupy Wall Street. We find that, compared to a network of stable domestic political communication, the Occupy Wall Street network exhibits higher levels of locality and a hub and spoke structure, in which the majority of non-local attention is allocated to high-profile locations such as New York, California, and Washington D.C. Moreover, we observe that information flows across state boundaries are more likely to contain framing language and references to the media, while communication among individuals in the same state is more likely to reference protest action and specific places and times. Tying these results to social movement theory, we propose that these features reflect the movement’s efforts to mobilize resources at the local level and to develop narrative frames that reinforce collective purpose at the national level.
Citation: Conover MD, Davis
C, Ferrara E, McKelvey K, Menczer F, et al. (2013) The Geospatial
Characteristics of a Social Movement Communication Network. PLoS ONE
8(3):
e55957.
doi:10.1371/journal.pone.0055957
Editor: Yamir Moreno, University of Zaragoza, Spain
Received: October 18, 2012; Accepted: January 4, 2013; Published: March 6, 2013
Copyright: © 2013 Conover et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The authors gratefully acknowledge support from the National Science Foundation [http://www.nsf.gov/] (grant CCF-1101743), DARPA [http://www.darpa.mil/] (grant W911NF-12-1-0037), and the McDonnell Foundation [http://www.jsmf.org/]. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the funding agencies. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Editor: Yamir Moreno, University of Zaragoza, Spain
Received: October 18, 2012; Accepted: January 4, 2013; Published: March 6, 2013
Copyright: © 2013 Conover et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The authors gratefully acknowledge support from the National Science Foundation [http://www.nsf.gov/] (grant CCF-1101743), DARPA [http://www.darpa.mil/] (grant W911NF-12-1-0037), and the McDonnell Foundation [http://www.jsmf.org/]. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the funding agencies. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
One of the most prominent American political movements of the past thirty years, Occupy Wall Street (‘Occupy’) is remarkable in the extent to which social media played a central role in its development and organization [1], [2]. In this study, we examine how the needs and constraints of social movements are reflected in the geospatial characteristics and information sharing practices of Twitter users engaged in communication about the Occupy movement. Specifically, we focus on the geographic distribution of these users and the ways in which the relationships among them diverge from those of users contributing to the two most popular streams for stable political discourse in the United States, ‘Top Conservatives on Twitter’ and ‘Progressives 2.0.’The organizing forces underlying successful social movements have been studied extensively by sociologists and political scientists [3]–[7]. From this body of work common themes have emerged, include the problems of resource mobilization and collective framing, which together constitute two of the core issues any social movement must address in order to effect social or political change. Resource mobilization refers to the process by which a social movement must marshal the financial, material, and human resources required to sustain its activities [8]. Collective framing is a process whereby the constituents of a social movement, through formal or informal processes, come to establish the narratives, language, and imagery that capture the essential features of the movement’s purpose and struggle [9]. Effective framing helps to foster a sense of community and engagement, and can be a powerful response to countervailing social pressures from establishment organizations [10].
Here we study Occupy Wall Street, a social movement focused on issues relating to the uneven distribution of wealth, social inequality, corporate greed, and the regulation of major financial institutions. Since the first protest on September 17
,
2011, a major feature of the movement has been the long-term physical
occupation of high-visibility encampments, often found in parks, banks,
libraries and foreclosed homes. As a result, the Occupy movement
requires substantial supporting infrastructure, including housing and
sanitation facilities, as well as access to communication technologies.
In spite of this, Occupy has sustained a lasting presence in American
cities including New York City, Oakland, Washington, D.C., and Boston,
which also represent key loci of decision making and protest activity [1], [2].
Under the Occupy model, proposals are brought to a vote before a
general assembly, a form of direct democracy in which any participant is
free to comment or vote on any proposal under consideration. The most
prominent among these organizational structures is the New York City
General Assembly, which has been responsible for producing policy and
key narrative frames such as the popular protest slogan, “We are the
99%,” which references the disproportionate concentration of wealth
among the top 1% of the world’s population [11].Social media have played a prominent role in facilitating communication and coordination throughout the development of the Occupy Wall Street movement. For example, the first call to action in the Canadian anticapitalist magazine ‘AdBusters’ used the Twitter ‘hashtag’ #occupywallstreet as one of just ten words featured in a full-page ad. Ever since, the Twitter platform has been used extensively by movement participants [2], with #ows being one of the hundred most popular hashtags on Twitter for the year 2011. In addition to Occupy, Twitter has also played a prominent role in several foreign social movements, most notably in the Egyptian revolutionary protests of 2011 [12]–[14].
In this work, we seek to understand the relationship between the geospatial dimensions of social movement communication networks and the organizational pressures facing such movements. To accomplish this, we use a state-of-the-art location inference technique to model relationships among users as a weighted directed network of communication flows between states, in which the weight of each edge corresponds to the volume of traffic between pairs of locations. Using this framework we investigate three distinct relationships: attention allocation and proximity to on-the-ground events, resource mobilization and localized information sharing, and the role of collective framing in long-distance communication.
With respect to the issue of attention allocation, we find that compared to stable domestic political communication the Occupy Wall Street movement exhibits very high levels of geographic concentration, with users in New York, California, and Washington D.C. producing more than half of all retweeted content. Aside from these hubs, however, we find that the appeal of content relating to Occupy Wall Street has a disproportionately local audience. With extended, high profile encampments and large-scale protest action playing central roles in the Occupy movement, we propose that this structural feature reflects the importance of mobilizing human resources at the local level.
Finally, we report on evidence indicating that the content of communication at the national level is distinct from the content of communication among users in the same state. Comparing intrastate versus interstate communication, we find that the terms most overrepresented in interstate communication relate to the movement’s core framing language and the news media, while the terms most overrepresented in local communication reference physical places, protest action, and specific times. These results support the hypothesis that local-level communication activity is driven by the challenge of resource mobilization, while long-distance communication is more strongly associated with collective framing processes.
Materials and Methods
Twitter Platform
Twitter is a popular social networking and microblogging site extensively explored in recent literature [15]–[21]. Among others, it has been used to study influence and credibility [22]–[26], social structure [27]–[29] and to monitor users’ sentiment [30]–[33]. Twitter users can post-character messages containing text and hyperlinks, called tweets,
and interact with one another in a variety of ways. Communication on
Twitter is characterized by directed, non-reciprocal social links that
allow users to subscribe to the stream of content produced by another
user. The content produced by every user an individual follows is
aggregated into a single streaming feed, from which an individual can
selectively rebroadcast content to his or her followers by choosing to
retweet it. In this way, a retweet serves to broaden the potential
audience of a piece of content, and signifies that information has been
transmitted between two individuals. Hashtags, short tokens prepended
with a pound sign (e.g. #taxes or #obama), constitute another important
feature of the platform, and allow the content produced by many
individuals to be aggregated into a custom, topic-specific stream
including all tweets containing a given token.Data
The analysis described in this article relies on data collected from the Twitter ‘gardenhose’ streaming API between July, 2011 and March ,
2012 – a nine month period including the birth and maturation of the
Occupy Wall Street movement. The gardenhose provides an approximately
sample of the entire Twitter stream in a machine-readable format.
Gardenhose tweets include useful metadata, among them a unique tweet
identifier, the content of the tweet (including hashtags and
hyperlinks), a timestamp, the username of the account that produced the
tweet, a free text ‘location’ string associated with the originating
user’s profile, and for retweets, the account names of the other users
associated with the tweet. Tweets from geolocation-enabled mobile
devices also report latitude/longitude coordinates, however the
incidence rate of tweets with this data is not enough to be useful as a
feature in general.To isolate a representative sample of Occupy Wall Street content we flagged for collection any tweet containing hashtags associated with the Occupy movement, including #ows and #occupy{*} (e.g. #occupywallst, #occupyboston, etc.). To provide a baseline against which to compare our observations, we also extracted content originating from the two most popular communication channels associated with stable domestic political communication, #tcot (Top Conservatives on Twitter) and #p2 (Progressives 2.0). In total, this sampling procedure produced 1,522,415 tweets associated with Occupy Wall Street and 825,262 tweets associated with domestic political communication. As this analysis is concerned primarily with information spreading processes we consider only retweet events from this corpus, resulting in 676,369 retweets among 257,657 users associated with Occupy Wall Street, and 259,703 retweets among 68,049 users associated with stable domestic political communication. Henceforth, we consider these corpora to constitute representative samples of retweet interactions among users participating in the streams of content associated with the Occupy Wall Street movement and stable domestic political communication in the United States.
Geocoding
To facilitate a geospatial analysis of communication activity associated with these content streams we require a high quality method to infer individual users’ locations. To accomplish this, we rely on self-reported location strings and the services of a commercial geocoding API. This technique, popularized in work by Onnela et al. [34], has been shown to produce high-resolution, high-quality geolocation data in the presence of geographically meaningful input.A caveat to this technique, however, is that it relies on raw text generated by a broad swath of the Twitter population, and so we find geographically meaningless location descriptors included in the dataset. To address this issue we rely on an extensive hand-curated blacklist of popular non-geographical responses such as ‘everywhere’ and ‘the dance floor’. To produce this list we sorted all location strings by popularity and reviewed the thousand most popular strings manually, blacklisting those that did not correspond to geographically meaningful entities. Drawn from a long tailed distribution, 53% of all tweets in the data set are associated with a location among the 1,000 most popular responses, with 27% of all tweets containing one of the top hundred location strings. From this set of one thousand we blacklisted 161 non-location strings, corresponding to 6% of the tweets associated with the 1,000 most popular responses.
To improve recall in the presence of novel input, we used a modified version of the Ratcliff-Obershelp algorithm [35] to detect fuzzy matches between free text location strings and the blacklist of popular non-location responses. As a result, because ‘the dance floor’ is in the set of blacklist responses, strings taking a slightly modified form, such as ‘on the dance floor,’ will also be classified as invalid input. The hand-coded blacklist combined with the Ratcliff-Obershelp fuzzy matching technique resulted in 9% of the free-text location strings being classified as non-location input.
From among the remaining responses we submitted location strings to the Bing.com geocoding API, which returns a best-guess estimate for the corresponding physical coordinates. This output is hierarchically formatted to describe the finest level of geographic resolution available. For example, if a user reports ‘Logan Square, Chicago’ as his or her location, the Bing API will return information about the likely zip code, city, state and country associated with that location. However, if the user reports only ‘USA,’ the information provided by the API describes only a country-level guess as to the user’s location. Owing to decreased coverage at the city-level and the proportionately few users associated with each individual city, we utilize the state-level location estimates for the geospatial components of this analysis.
In total, 68.4% of Occupy Wall Street users reported location strings, and from these we were able to obtain geolocation estimates for 55.7% of these accounts. Among this set of users, 60% of the resulting geolocation estimates included state-level metadata. Response rates were somewhat diminished for users associated with the stream of domestic political communication, with 36% of individuals reporting free-text location strings. Using the procedure described above, we were able to obtain geolocation estimates for 29.3% of all users in the domestic political communication stream, 82.4% of which contained state-level metadata.
Geographic Profile
One of our goals is to establish a coarse-grained geographic profile for communication activity associated with the Occupy Wall Street movement. Formally, for each stream we define an activity distribution across states as,, where is the total number of retweets originating from state and
is the total number of retweets originating from all states. As
outlined above, we focus on retweets as they correspond to attention
allocation rather than total content production volume.In addition to the distribution of activity across individual states we examine the information sharing relationships among users in different locations. To accomplish this, we rely on a network representation to characterize the flow of information on Twitter. Taking users as nodes, we define a weighted directed network in which an edge with weight
is drawn from node to in the event that user retweets user
times. The intuition underlying this approach is that each retweet
provides evidence suggesting that information produced by user was evaluated and acted upon by user .Combining the user-level geocode metadata previously described with the network representation defined here we can induce another network describing the volume of communication between users in each state. In this network, nodes represent states, and weighted directed edges are drawn among them. The weight of the edge from
to is defined as the sum of the weights among all edges originating from users in state and terminating in state .
We note, however, that this induced network must have geolocation
labels for each node in a dyad. In the Occupy Wall Street stream we
identify 143,437 tweets for which both the source and target have
state-level geolocation data and 78,467 likewise restricted tweets in
the stream of stable domestic political communication.Textual Content
Finally, we wish to investigate whether the content of tweets with different geospatial properties serve distinct communication functions. To accomplish this, we segregate Occupy Wall Street tweets into two classes: interstate tweets connect pairs of users in different states, and intrastate tweets connect users in the same state. We compute the probability of observing a token,, in a tweet from a given class, , as . Comparing these probabilities yields a ratio, ,
a value which is large when a token is more common in intrastate
traffic than interstate traffic and small under the opposite conditions.Results
Geographic Concentration
Figure 1, in which states are ordered according to the proportion of stream activity, shows that content in the Occupy stream is substantially more geographically concentrated in a few key states compared to domestic political communication. For example, New York accounts for 30% of the total retweet activity in the Occupy stream, while the most popular source for stable domestic political communication, Washington D.C., accounts for only 10.7% of the stream’s total volume. As these plots make clear, the primary locations for on-the-ground Occupy activity are those places responsible for the majority of widely rebroadcast Occupy content, with California, New York and Washington D.C. acting as the source of 53.8% of total retweets. Figure 2 maps the states where the proportion of activity associated with the Occupy stream deviates the most from that associated with the stream of domestic political communication.Ordered by the amount of traffic associated with each state, it is clear that a few high-profile locations serve as the dominant sources of content in the Occupy stream. This concentration stands in contrast to the more heterogeneous activity profile for the stream of domestic political communication.
doi:10.1371/journal.pone.0055957.g001
This cartogram uses color to represent the extent to which the number of Occupy Wall Street tweets in each state deviates from the domestic political communication baseline, computed as:
.
Redder colors indicate that proportionally more Occupy content
originated from the associated state, while whiter colors indicate the
opposite. To minimize the effect of outliers on the visualization and to
highlight variation between states, colors for Maine and Oregon have
been fixed, indicating that the deviation from baseline is more than
three times the expected rate.doi:10.1371/journal.pone.0055957.g002
Occupy Wall Street users, by state, exhibit a lower content production to consumption ratio relative to users in the domestic political communication stream. The disproportionately high ratio observed for Kentucky can be attributed to the activity of a prolific, highly popular left-leaning user from that state.
doi:10.1371/journal.pone.0055957.g003
,
this technique allows for the selection of statistically significant
edges across all weight scales, a feature that is especially valuable
when working with networks with heterogeneous weight distributions such
as those associated with communication or human mobility.Applying this technique to both networks reveals a communication backbone for the Occupy network that exhibits the highly concentrated hub and spoke structure described above. Figure 4 shows that the Occupy Wall Street network is characterized by minimal state-to-state connectivity, with the majority of statistically significant traffic flowing to and from New York, California and Washington D.C. This is in contrast to the communication backbone for the network of domestic political communication, in which we observe extensive interactions among many pairs of states.
) of the continental interstate communication networks.Stable domestic political communication is shown at left, Occupy Wall Street at right. Edges adjacent to New York, California, and Washington D.C. are shown in red. Note that Occupy Wall Street exhibits a clear hub-and-spoke pattern, with the majority of traffic flowing to and from New York, California, and Washington, D.C. Likewise, observe that the Occupy Wall Street network exhibits diminished levels of interstate connectivity compared to the network of domestic political communication. We note that the structure of this network backbone is robust to different parameterizations of alpha.
doi:10.1371/journal.pone.0055957.g004
Localization
In Figure 5 we present interstate connectivity for each communication network as a matrix in which the weight of an edge is mapped to a grayscale hue ranging from white for weak relationships to black for the strongest relationships. Inspecting these plots, one of the most striking ways in which the topology of the Occupy Wall Street communication network departs from that of the domestic political communication network is the high degree of localization. This is evidenced by the presence of a strong diagonal in the Occupy Wall Street connectivity matrix, as well as the significant off-diagonal mass in the domestic political communication matrix. We find that 40% of Occupy retweets originate and terminate with users in the same state. In contrast, 11% of retweets from the domestic political stream exhibit this type of locality, an increase of more than 350%.The edge weight corresponding to each cell is mapped to a color hue on a logarithmic scale ranging from white for edges with the least weight to black for edges with the most weight. The strong diaonalization and limited off-diagonal mass apparent in the Occupy Wall Street matrix is indicative of highly localized communication activity.
doi:10.1371/journal.pone.0055957.g005
Textual Analysis
To study the relationship between geography, resource mobilization, and collective framing, we focus on the content of tweets flowing within and between states. Restricting our analysis to tokens that account for at least 0.1% of both the intrastate and interstate tweet text, Table 1 presents the ten tokens most overrepresented in both intrastate communication as well as interstate communication.doi:10.1371/journal.pone.0055957.t001
Discussion
The analysis of interstate connectivity patterns reveals that, relative to stable domestic political communication, the Occupy network has a highly localized geospatial structure, with a disproportionately large amount of traffic being produced and consumed by users in the same state. We propose that this phenomenon may be related to the issue of resource mobilization, that is, the process whereby any social movement must marshall resources such as money, infrastructure and human capital to further the goals of the movement. In the case of Occupy Wall Street, such resources are often quite tangible, and include not only tents and food, but also the participants required to facilitate large-scale protest action and extended encampments in cities across the country. In this light, it is easy to understand why such a disproportionately large fraction of attention is allocated to communication at the local level.With respect to the finding that the majority of widely rebroadcast content is produced by users in a small number of high profile locations, we observe that these states represent sites of major encampment and decision making activity. Despite the fact that all users can contribute equally to the Occupy stream, it appears that proximity to events on the ground plays a major role in determining which content receives the most attention. This is in contrast to the stream of domestic political communication, in which content from users across the United States is allocated a significant share of attention. Where the stream of domestic political communication looks more like a conversation taking place at the national level, the structure of the Occupy stream is more akin to a broadcast, with just a few locations playing the role of net content producers.
Finally, we propose that interstate communication plays a significant role in the propagation of narrative imagery associated with collective framing processes, and that intrastate communication is driven more predominantly by the pressures of resource mobilization. Looking to the lists of tokens most overrepresented in each type of traffic (Table 1), we find that those more common in interstate communication include references to core framing language and the news media. This finding suggests that when users engage in communication across state boundaries they allocate proportionately higher levels of attention to speech associated with collective framing processes. In contrast, the tokens more common in intrastate traffic relate to protest action and specific times and places. From this we conclude that the content of intrastate tweets deals much more frequently with rallying the movement’s participants, a core function of resource mobilization.
The findings outlined in this paper dovetail nicely with established literature on social movement theory. However, statistical measures are limited in the extent to which they can accurately represent nuanced features of communication, and future work in this domain would benefit from rigorous qualitative content analysis. Moreover, there remains room to improve our understanding of how closely the structure of social media communication mirrors that of other forms of communication. For example, Mislove, et al. found that the geographical distribution of Twitter users tends to over-represent populous counties and metropolitan areas, suggesting that entire rural regions may be significantly under-represented – with similar findings holding true for ethnicity and gender as well [37]. In this respect as well, work of this nature would benefit from deeper involvement from scholars in the social sciences, and we hope that this type of interdisciplinary collaboration will become increasingly common.
Acknowledgments
We would like to thank Alex Rudnick, Jacob Ratkiewicz, Mark Meiss, and other current and past members of the Truthy group at Indiana University (cnets.indiana.edu/groups/nan/truthy) for their contributions to the Truthy Project. Additionally, we would like to thank Fabio Rojas, Brian Keegan, and Bruno Gonçalves for their constructive, insightful feedback during the preparation of this manuscript.Author Contributions
Conceived and designed the experiments: MDC AF FM EF CD KM. Performed the experiments: MDC CD. Analyzed the data: MDC AF FM EF CD KM. Wrote the paper: MDC AF FM EF.References
- 1. Chomsky N (2012) Occupy. Zuccotti Park Press.
- 2. Byrne J (2012) Occupy the media: Journalism for (and by) the 99 percent. In: Bryne J, editor, The Occupy Handbook. Little, Brown, 256–264.
- 3. Myers D (1994) Communication technology and social movements: Contributions of computer networks to activism. Social Science Computer Review 12: 250–260. doi: 10.1177/089443939401200209. Find this article online
- 4. Van Laer J, Van Aelst P (2009) Cyber-protest and civil society: the internet and action repertoires in social movements. Handbook on Internet Crime : 230–254.
- 5. Wray S (1999) On electronic civil disobedience. Peace Review 11: 107–111. doi: 10.1080/10402659908426237. Find this article online
- 6. Bennett W (2007) Changing citizenship in the digital age. The John D and Catherine T MacArthur Foundation Series on Digital Media and Learning : 1–24.
- 7. Bennett W (2003) Communicating global activism: Strength and vulnerabilities of networked politics. Information, Communication & Society 6: 143–168. Find this article online
- 8. McCarthy J, Zald M (1977) Resource mobilization and social movements: A partial theory. American Journal of Sociology: 1212–1241.
- 9. Benford R, Snow D (2000) Framing processes and social movements: An overview and assessment. Annual Review of Sociology : 611–639.
- 10. Garrett R (2006) Protest in an information society: A review of literature on social movements and new ICTs. Information, Communication & Society 9: 202–224. Find this article online
- 11. MotherJones website (Accessed 20 Jan 2013). “We Are the 99 Percent” Creators Revealed. URL http://www.motherjones.com/politics/2011/10/we-are-the-99-percent-creators.
- 12. Choudhary A, Hendrix W, Lee K, Palsetia D, Liao W (2012) Social media evolution of the egyptian revolution. Communications of the ACM 55: 74–80. doi: 10.1145/2160718.2160736. Find this article online
- 13. Lotan G, Graeff E, Ananny M, Gaffney D, Pearce I, et al.. (2011) The revolutions were tweeted: Information flows during the 2011 tunisian and egyptian revolutions. International Journal of Communication 5.
- 14. González-Bailón S, Borge-Holthoefer J, Rivero A, Moreno Y (2011) The dynamics of protest recruitment through an online network. Scientific reports 1.
- 15. Java A, Song X, Finin T, Tseng B (2007) Why we Twitter: understanding microblogging usage and communities. In: Proceedings of the 9th WebKDD and 1st SNA-KDD 2007 workshop on Web mining and social network analysis. ACM, 56–65.
- 16. Krishnamurthy B, Gill P, Arlitt M (2008) A few chirps about Twitter. In: Proceedings of the 1st workshop on Online social networks. ACM, 19–24.
- 17. Huberman B, Romero D, Wu F (2009) Social networks that matter: Twitter under the microscope. First Monday 14: 1–5. doi: 10.2139/ssrn.1313405. Find this article online
- 18. Jansen B, Zhang M, Sobel K, Chowdury A (2009) Twitter power: Tweets as electronic word of mouth. Journal of the American society for information science and technology 60: 2169–2188. doi: 10.1002/asi.21149. Find this article online
- 19. Kwak H, Lee C, Park H, Moon S (2010) What is Twitter, a social network or a news media? In: Proceedings of the 19th International Conference on World Wide Web. ACM, 591–600.
- 20. Boyd D, Golder S, Lotan G (2010) Tweet, tweet, retweet: Conversational aspects of retweeting on Twitter. In: Proceedings of the 43rd Hawaii International Conference on System Sciences. IEEE, 1–10.
- 21. Wu S, Hofman J, Mason W, Watts D (2011) Who says what to whom on Twitter. In: Proceedings of the 20th International Conference on World Wide Web. ACM, 705–714.
- 22. Cha M, Haddadi H, Benevenuto F, Gummadi K (2010) Measuring user influence in Twitter: The million follower fallacy. In: Proceedings of the 4th International AAAI Conference on Weblogs and Social Media.
- 23. Bakshy E, Hofman J, Mason W, Watts D (2011) Everyone’s an influencer: quantifying influence on Twitter. In: Proceedings of the 4th ACM international conference on Web search and data mining. ACM, 65–74.
- 24. Castillo C, Mendoza M, Poblete B (2011) Information credibility on Twitter. In: Proceedings of the 20th International Conference on World Wide Web. ACM, 675–684.
- 25. Frahm K, Shepelyansky D (2012) Google matrix of Twitter. The European Physical Journal B 85: 1–7. doi: 10.1140/epjb/e2012-30599-6. Find this article online
- 26. Weng L, Flammini A, Vespignani A, Menczer F (2012) Competition among memes in a world with limited attention. Scientific Reports 2.
- 27. Conover M, Ratkiewicz J, Francisco M, Gonçalves B, Flammini A, et al.. (2011) Political polarization on Twitter. Proceedings of the 5th International AAAI Conference on Weblogs and Social Media.
- 28. Grabowicz P, Ramasco J, Moro E, Pujol J, Eguiluz V (2012) Social features of online networks: The strength of intermediary ties in online social media. PloS one 7: e29358. doi: 10.1371/journal.pone.0029358. Find this article online
- 29. Mathiesen J, Yde P, Jensen MH (2012) Modular networks of word correlations on Twitter. Scientific Reports 2.
- 30. Tumasjan A, Sprenger T, Sandner P, Welpe I (2010) Predicting elections with Twitter: What 140 characters reveal about political sentiment. In: Proceedings of the 4th International AAAI Conference on Weblogs and Social Media. 178–185.
- 31. Bollen J, Mao H, Zeng X (2011) Twitter mood predicts the stock market. Journal of Computational Science 2: 1–8. doi: 10.1016/j.jocs.2010.12.007. Find this article online
- 32. De Choudhury M, Counts S, Gamon M (2012) Not all moods are created equal! exploring human emotional states in social media. In: Proceedings of the 6th International AAAI Conference on Weblogs and Social Media.
- 33. Ciulla F, Mocanu D, Baronchelli A, Goncalves B, Perra N, et al. (2012) Beating the news using social media: the case study of american idol. EPJ Data Science 1: 8. doi: 10.1140/epjds8. Find this article online
- 34. Onnela J, Arbesman S, González M, Barabási A, Christakis N (2011) Geographic constraints on social network groups.
- 35. Ratcliff J, Metzener D, et al. (1988) Pattern matching: The gestalt approach. Dr Dobb’s Journal 7: 46. Find this article online
- 36. Serrano M, Boguñá M, Vespignani A (2009) Extracting the multiscale backbone of complex weighted networks. Proceedings of the National Academy of Sciences 106: 6483–6488. doi: 10.1073/pnas.0808904106. Find this article online
- 37. Mislove A, Lehmann S, Ahn Y, Onnela J, Rosenquist J (2011) Understanding the demographics of Twitter users. In: Proceedings of the 5th International AAAI Conference on Weblogs and Social Media
Para seguir o blog e receber postagens atualizadas, use a opção "seguir", ao lado.
DEPUTADO do DEM quer censura,propondo facilidades para remover conteudos relacionados a parlamentares http://t.co/BQ8zRUBSaT
ResponderExcluirA INTERNET ajudou a mobilizar todos os egicios na praça Tanrir contra HOSNI MOBARAK
ResponderExcluir