CC BY-NC-ND
0Available online at www.sciencedirect.com
ScienceDirect
Procedia - Social and Behavioral Sciences 100 (2013) 40 - 56
8th Conference on Applications of Social Network Analysis - ASNA 2011
Mutual influence of twitter and postelection events of Iranian
presidential election
Kaveh Ketabchia, Masoud Asadpourab, Seyed Amin Tabatabaeia
aSocial Networks Lab,Faculty of Electrical and Computer Engineering, University ofTehran, Tehran, Iran bSchool ofComputer Science, Institute forResearch in Fundamental Sciences (IPM), P.O.Box 19395-5746, Tehran, Iran
Abstract
The 10th Iranian presidential election, held on 12 June 2009, was the most important political event in Iran during the last decade. Postelection protests were raised after announcement of the results. Social networking web sites including Twitter and Facebook played an important role in distribution of news, images and videos of the events in that period of time. It is believed that these social networks were quite important for protesters to organize and manage their demonstrations.
In this research, we study the tweets published in Twitter around this subject, from 3 months before the election until 15 months after the election. We study the structure of the social network, its dynamics and the mutual effects of Twitter and Iranian people on post-election protests and events.
We show that the most active users were joined just a few days after the election when some mobile services were slowed-down by the telecommunication companies. We also show that Tweets were mainly used to communicate the events to the outside world, i.e. the Twitter was used mainly as a one-way media.
We also analyze the content of tweets and find the most frequent words and patterns. We show how the events affect the rate of tweet publication and how famous politicians in real society affect tweets in virtual society.
© 2013TheAuthors.PublishedbyElsevierLtd.
Selection and/or peer-review under responsibility of Dr. Manuel Fischer
Keywords: Social Network Analysis; Iranian Presidential Election; Twitter
Corresponding author: asadpour@ut.ac.ir
1877-0428 © 2013 The Authors. Published by Elsevier Ltd. Selection and/or peer-review under responsibility of Dr. Manuel Fischer doi: 10.1016/j.sbspro.2013.10.698
1. Introduction
Twitter1 is a website, which offers a social networking and micro-blogging service, and enables users to send and receive messages called tweets. Tweets are text-based posts of up to 140 characters displayed on a user's profile page. Many works have already been done to analyze and investigate social networks in different countries and communities or in specific events. Adamic & Glance (2005) had studied posts of several webblogs and analyzed their impact on 2004 United States presidential elections. Some works have studied Arabic (Etling, Kelly, Faris, & Palfrey, 2009) and Persian (Kelly & Etling, 2008; Qazvinian, Rassolian, Shafei, & Adibi, 2007) blogosphere . The role of social networks in the Arabic Spring movements has also been studied (Ghannam, 2011; Cottle, 2011; Rane & Salem, 2012). The 2009 Iran's presidential election was the most important political event in Iran after the Islamic revolution in 1978. It is believed that social networking websites like Twitter played an important role in postelection protests after election (12 June 2009). Twitter users found the Iranian elections the most engaging topic in 2009. The terms #iranelection, Iran and Tehran were all in the top-21 of Trending Topics2. It is believed that social networking web sites were used mostly for advertisement and organization of protests and distribution of news, images and videos of those events (Khonsari, Nayeri, Fathalian, & Fathalian, 2010).
In this research we study the tweets published in Twitter around this subject, from 3 months before the election until 15 months after the election. We study the structure of the social network, its dynamics and the mutual effects of Twitter and Iranian people on post-election protests and events.
2. Data collection
Twitter provides an API3 which facilitates accessing users' tweets and friends. To find the active users in Iran election event, a list of related keywords was collected. This list contained both Persian and English keywords. It was grouped in five categories: politicians' name, protest days, active parties, and some keywords related to the word "demonstration" in Persian. We searched theses keywords in Twitter and collected the list of users who used at least one of the keywords in a period of 3 months before the election and 15 months after it (totally 18 months). This preliminary list contained 9,942 users. Then, we collected their profiles, tweets, and followers/followees up to the limits allowed by Twitter (Twitter's API returns up to 3200 most recent tweets and we started collecting data from 22 June 2010). In total 2,029,108 user profiles and 7,518,315 tweets were retrieved. Twitter provided access to the following information about users: id, name, number of followers, number of friends and account creation date. Also the following properties could be retrieved for tweets: id, user (owner), body text, creation date.
Besides the above information, some information could be retrieved through parsing the tweet's content e.g. retweeting or mentioning. Some users forward received tweets to their followers. This action is popularly known as retweeting and can be identified by the use of "RT @username" in the tweets. Also, users can respond to or comment on other people's tweets. This is called by Twitter as mentioning. Mentioning is identified by searching for "@username" in the tweet content, after excluding retweets.
1 http://twitter.com
2 Top Twitter Trends of 2009. The Official Twitter Blog. http://blog.twitter.com/2009/12/top-twitter-trends-of-2009.html.
3 Twitter Developers - API Documentation, https://dev.twitter.com/docs , accessed on 30/07/2011
3. Structural analysis of the social network
In this section, we give a summary of the structural analysis we did on the processed dataset. Some structural analysis like centrality indices are not presented here as they go into more details and might not be interesting for a general reader.
3.1. Scale-free property
Twitter as a friendship network can be analyzed by inspecting the degree distributions of users. Figure 1 and 2 show the histogram of in-degree (number of followers) and out-degree (number of followees) of users in log-log scale. As the figures demonstrate, the histogram is almost linear with a heavy tail. This indicates that both in and out-degrees have power-law distribution and the network is scale-free (Barabasi & Albert, 1999). R2 in the figures equals the square of the correlation coefficient between the observed and modeled data values. As the figures demonstrate, there is a high correlation coefficient between the linear model and data.
Fig. 1. Histogram of followers in log-scale
x y = 80684X 1375 R2 = 0.9119
• •
1 10 loo lOOO lOOOO
Log(out-degree)
Fig. 2. Histogram of followees in log-scale
3.2. Networkgrowth
We studied the evolution of the activity of Iranian users interested in presidential election in order to find whether election and protests have been influential in this regard or not. Figure 3 shows how many users have joined Twitter on specific months before and after the election.
The figure clearly shows that most users signed up in March (beginning of the new year in Persian calendar), April, May and June (month of election), 2009. On May 23, 2009, Iranian government started filtering Twitter. That might be why the number of new users in this month is a bit smaller than April as newcomers did not know how to use anti-filtering software.
The number of new users reached its peak at June and then started declining until the next March and reached a negligible number. Note that the figure does not mean the Iranian users are not interested anymore to Twitter. Since we have focused only on the tweets about presidential election and postelection events, the figure means only these issues are not interesting anymore for the Iranian community not the whole Twitter.
Fig. 3. Number of new users per month
In order to find the reason behindjoining in Twitter from Iran, we took a closer look at the daily rate of sign-ups. Figure 4 shows that, starting from the day of election, users join Twitter with an accelerating rate until four days after the election. The acceleration might be due to the fact that text-messaging services were down on mobile networks during the day of election. So people started using Twitter along with other social networking sites, like Facebook, to send news about election to the outside world. Micro-blogging services provided a fast way for protesters to share their observation and information and potentially to organize the next protests.
The largest peak of the diagram corresponds to the mass rally of protestors on June 14. After this day, the rate of new users suddenly dropped until 19th. On Friday, June 19, 2009, which was a weekend in Iran, Ayatollah Khamenei (The supreme leader) made a hardline speech at Friday prayers. On Saturday (the first day of the week in Persian calendar), June 20, the new users increased again. On this day,
opposition movement continued their protests, in response to the invitation of two defeated candidates, Mousavi and Karroubi. The other crucial event this day was a meeting of Iran's powerful guardian council, which had invited the three defeated candidates to voice their complaints. Then, the number of new users declined more and more.
Fig. 4. Number of new users per day for June, 2009
3.3. User activities
To measure the activity of users in this network on different days, the total number of tweets per day was calculated. Figure 5 shows the results. Since number of users increased over time it is clear that the number of tweets grows gradually. However, peaks of the graph shows important days, which correspond to critical political events. Among them, the marked ones will be explained below.
Fig. 5. Number of Tweets per day
Tweet publication rate started to increase on June 12 (Iranian presidential Election Day) and reached its maximum on June 20 (speech of the supreme leader in Friday prayers). The first week after the Iranian presidential election was the most prolific period for protesters. The 2nd peak corresponds to the rally in memory of student protests in July 9, 1999. The 3rd peak is associated with Friday prayers by Hashemi Rafsanjani. The 4th peak corresponds to Qods day rally on September 18. Although it was an annual rally in support of Palestinian people, protesters came to streets and made their objections to the government crackdown. The 5th peak corresponds to the Student's day rally held on November 4. One month later, on Scholar's day, university students held a protest against the government policies (6th peak of diagram). On Ashura which is a religious event, there was a rally in support of leaders of green movement which finally was led to violence (7th peak). On February 11th, there was a mass rally in support of government in which protesters failed to show their disagreement with a crackdown (8th peak).
Figure 5 showed the total activity of users. To find which users were more active, we had to calculate the average number of tweets per user. We therefore calculated the average number of daily tweets that users had sent from the beginning of their membership until the last day of data collection. Then we grouped them based on their subscription start month. Our analysis, demonstrated in fig. 6, show that the most active users have started their membership from June 2009 (the month of election), until December 2009, when large pro-government demonstrations started raising.
Fig. 6. Average number of tweets per day for each user based on membership start date 3.4. Follower-followee network
Users can be looked at in a social network based on follower-followee relationships. When node A follows the tweets of node B, we add an arc from A to B. In order to examine follower-followee network structure, it has been drawn using the force atlas layout of Gephi (Bastian M., 2009) open-source software. This layout runs a force-based algorithm that has two principles in drawing. First, vertices connected by an edge should be placed near each other. Second, vertices should not be placed too close to each other. Colors have been assigned to nodes based on users' language. English speaking users are colored in red, English and Persian speaking users are colored in green and Persian speaking users are colored in blue. In addition, node size is logarithmically proportional to the number of user's followers.
In order to find out the users' language we detect the language of tweets. Based on the character codes, we find out whether the tweet is in Persian or in English (other languages are not collected in our dataset). Users that have written all their tweets in English (Persian) are considered as English (Persian)-speaking user. The users that have written some (at least one) tweets in Persian and some in English are considered as Persian-English speaking user (the Persian-English speaking users are usually Persian natives who can speak English; the reverse case is rare).
In order to increase the visibility of the graph, some isolated nodes (less than 50 nodes) have been removed. The result is shown in fig. 7 which consists of one giant component with 4 small sub-communities: a small sub-community of English-speaking users on top-left of the giant component and two small communities of Persian-English speaking users on top and bottom-left and a big subcommunity in the center. The Persian speaking users (blue circles) are almost uniformly distributed in the giant component.
Fig. 7. follower-followee network of users. English speaking users are colored in red, Persian-English speaking users are colored in green and only Persian speaking users are colored in blue.
Almost all of the biggest nodes are located in the periphery. This is due to the fact that most of their followers are absent in the graph. Therefore they are moved toward the periphery by the drawing algorithm. These nodes act like a transmitter. They relay tweets from this community to the others. Nodes that are close to the center of the communities, have built a dense network among themselves and they are mostly among Persian-English speaking users.
3.5. Reply network
One way of measuring user's influence is by analyzing the replies. Replying indicates stronger interaction between users than follower-followee relationship. We draw the reply network by adding an edge from node A to node B if node A has replied to at least one tweet of node B. Figure 8 shows the results. The network has been visualized using Gephi (Bastian M., 2009) and the layout used here is called "Yifan Hu" (Hu, 2005). This layout is again among force-based algorithms for visualization. Two pictures of the same graph is seen, the left one is colored based on user-language and the right one is colored based on membership date. The network contains a giant component consisting of two large communities with lots of links within and few links between them.
The left figure shows that, users who speak both in Persian and English (colored in yellow) are situated in the center of both communities. The bigger community consists of many English speaking users (colored in blue) that are situated mostly in the periphery of the community. The smaller community consists mostly of Persian-English speaking users. Users in this community are Persian natives who can speak English as a second language. It is hard to find but there exists a few nodes that speak only in Persian in their tweets (colored in red).
Fig. 8. Reply network, node size is proportional to the number of followers, left) nodes are colored based on user language: English (blue), Persian (red), Persian-English (yellow), Right) The same network but nodes are colored based on their membership date in Twitter: before election (red), after election (blue).
The right figure shows communities with different colors. If we look at the center of the two communities we would see the center of the left community (the smaller one) consists mainly of users that have joined Twitter before election (colored in red). However, in the right community (the bigger one) we see the center consists mainly of users that have joined Twitter after the election (colored in blue).
3.6. Most active-users' network
We find the network among the most active, or in another word, most-prolific user. We call a user an active user if he has published more than 500 tweets in the period of observation (roughly 2 tweets per day during the 18 months period). Direction of links in this network means "the source node follows the status of the destination node". The difference between out-degree and in-degree in this network shows whether an active user behaves mainly as an influencer or as a conductor. Users with mostly incoming links are followed by many users and thus make one-way influence. However, the ones with mainly outgoing links, while following some active users, relay their messages to not-so-active users.
In fig. 9, we try to visualize the network of active users by placing the influencers near the top and the conductors near the bottom. The most active sender is "Mousavil388". This id belongs to the campaign of one of the candidates, Mir Hoseyn Mousavi. The most active transmitter is "IranElection" that forwards tweets of active users to the other users.
Fig. 9. Network among the most active-users
4. Content analysis of tweets
Many works have already been done to analyze and investigate the tweets' content for different purposes. Owen et al. (2009) describe a new approach to recommend news stories to the Twitter users. This technique harness real-time twitter data as the basis for ranking and recommending articles from a collection of RSS feeds. Hannon, Bennett, & Smyth (2010) focus on one of the key features of the social web. The authors of that article proposed that Twitter users can be modeled by their tweets and relationships of their Twitter social graph. They have demonstrated how these profiles can be used as the basis for a followee recommender. In this section we try to answer some questions about tweets contents such as: what types of contents were popular in tweets about Iran election? Which words have had the most usage in tweets? Which patterns (set of words) had the most usage? Which politicians had been mentioned the most? Why a particular politician had been mentioned very much in a specific day?
To answer these questions, content of tweets containing "#iranelection" tag were analyzed. In order to do a thorough investigation, this specific set has been broken down into three subsets including:
• P tweets: Tweets which have been written in Persian (about 270,000 tweets, 20%).
• E tweets: tweets which have been written in English by foreigners. Precisely speaking, if a user has no tweet in Persian we consider him/her as a foreigner and treat his/her tweets as an E tweet (about 150,000 tweets, 11%).
• PE tweets: Tweets which have been written in English by Persian speaking natives i.e. those that have had at least one tweet in Persian (about 940,000 tweets, 69%).
Then, the most common words and most frequent patterns in each set of tweets were found.
4.1. Frequent words
We tried to find keywords which had been used frequently in tweets. First, we removed the stopwords because they are not informative. We used the set of Persian stopwords provided by (AleAhmad, Amiri, Darrudi, Rahgozar, & Oroumchian, 2009) in Hamshahri corpus. For English stopwords the set from Reuters21578 corpus (Lewis, 1999) was used. Meanwhile, we found that some users had misused the #iranelection tag and had sent spam. Therefore, we had to filter out some frequent words that were used by that small number of users.
Tables 1 to 3 show the 40 most frequent words in P, E, and PE Tweets respectively. Some interesting conclusion can be made by comparing the tables:
• Among the news agencies, BBC has appeared more in the tweets of Persian speaking users since it has a dedicated channel in Persian (BBC Persian) which was very active those days. However, for English speaking users, CNN has been the most cited news agency.
• For Persian speaking users, the most influential persons have been: Mousavi, Khamenei, Karroubi, Ahmadinejad, respectively. For English speaking users the most influential ones have been: Neda (Aqa Soltan), Mousavi, Ahmadinejad, Khamenei, and Obama, respectively.
• Most protests were held in Tehran by Scholars near the University of Tehran so that is why "Tehran", "University", and "Scholars" are among the mostly used words.
• Persian Tweets have been reporting rough words e.g. arrest, prison(ers), execution, force, and protest: while English tweets are full of sympathy words e.g. support, show, free, and help.
• Persian users that have written English tweets, besides the words that are used mainly for live reporting the events (e.g. today, now,just), used some imperative words (e.g. please, don't). These words were used mainly for guiding the international media for better coverage of the events.
Kaveh Ketabchi et al. /Procedia - Social and Behavioral Sciences 100 (2013) 40 - 56 Table 1. The most frequent words in P Tweets
order Word Translation order Word Translation
1 иЫ Iran 21 Scholars
2 People 22 Movement
3 Tehran 23 AyatoHah
4 Regime 24 Revolution
5 Green 25 Today
6 University 26 Wednesday
7 Freedom 27 Government
8 Mousavi 28 iljl Free
9 Khamenei 29 oM^J J Prisoners
10 <JJI Allah 30 Force
11 Arrest 31 Slogan
12 Prison 32 оЬЦ> Street
13 Political 33 Velayat
14 Execution 34 Square
15 Karroubi 35 olj Way
16 Ahmadinejad 36 Photo
17 iSj-O Death 37 Protest
18 JU Year 38 Family
19 BBC 39 J-^j Generation
20 Islamic 40 Republic
Table 2. The most frequent words in E Tweets
Order Word Order Word
1 Iran 21 Freedom
2 Tehran 22 Ahmadinejad
3 Neda 23 Today
4 GR88 24 Protest
5 Iranian 25 Irans
6 People 26 Show
7 Now 27 Democracy
8 Green 28 Free
9 News 29 Khamenei
10 Time 30 Iranians
11 Please 31 CNN
12 Twitter 32 Protests
13 Mousavi 33 Revolution
14 Support 34 Help
15 US 35 Protesters
16 New 36 Topix
17 World 37 Persian
18 Video 38 Basij
19 Regime 39 Police
20 Ppl 40 Obama
Kaveh Ketabchi et al. / Procedia - Social and Behavioral Sciences 100 (2013) 40 - 56 Table 3. The most frequent words in PE tweets
Order Word Order Word
1 Iran 21 Day
2 Tehran 22 Arrested
3 Neda 23 Please
4 GR88 24 Protests
5 Iranian 25 World
6 People 26 Death
7 News 27 Irans
8 Ppl 28 Khamenei
9 Green 29 Don't
10 Mousavi 30 Free
11 Video 31 Prison
12 Regime 32 Ahmadinejad
13 Now 33 Time
14 IR 34 Police
15 US 35 Support
16 Freedom 36 Protesters
17 Protest 37 Iranians
18 Today 38 BBC
19 New 39 Basij
20 CNN 40 Right
Table 4. The top most frequent patterns of words
Pattern Translation Number of tweets Number of users
B. B. c. 8399 739
Ahmadi Nejad 5757 577
aJUl Ayat Allah 5213 511
J З^-Я.ino Masoud Rajavi 4417 95
University Student 4321 461
J-^-i Young generation 4081 54
Green movement 3926 497
Islamic Republic 3277 433
Political prisoners 3164 349
Mir Hosey Mousavi 2746 467
Velayat-e faqih 2695 323
4.2. Frequent patterns
We also found the frequent patterns of words (Beil et al, 2009) which were used the most in P tweets. Patterns are set of words that come together in one tweet. Table 4 shows the top used patterns along with the number of tweets and the number of users that had used them. The top pattern is "BBC". This word is written in 3 separate tokens in Persian (as it is pronounced). That is why it is considered as a pattern. The next pattern, "Ahmadi nejad" is the last name of the president, related to almost all events. The third pattern, "Ayat-allah", is a title for clerics. Many famous cleric politicians e.g. Khamenei, Hashemi, and Montazeri have this title; therefore lots of tweets containing their name consist of this pattern. The other patterns have their reason to be on the list of top patterns e.g. most protests were held by "University students" in support of "Green Movement". It is interesting that "Green Movement" is repeated in more tweets than "Mir Hoseyn Mousavi" who was leading it.
Among all patterns, two of them have been used only by a small group of users (i.e. "Masoud Rajavi" and "Young generation"). These patterns were used by a group known in Iran as an armed terrorist group called Mojahedin (called Monafeqin in Iran) led by Masoud Rajavi. They were trying to benefit from the situation and attract people toward their group.
4.3. Number of tweets per day
In this part, we focus on the number of tweets sent per day and try to understand why and when in some days users were more active. Also, we study the tweets that mention famous politicians. Figure 10 shows the number of tweets that contain #Iranelection tag, in three sets of tweets. As one can see, in early days after the election, the number of English tweets which had been written by foreign users (E tweets) was more than the number of tweets in other sets. That was due to the interests of foreign users to the events of Iran specially the death of Neda Aqa Soltan. However, this order changed two months after and foreign users lost their interest in following events of Iran.
Except for some days just after the election until 3 days after the death of Neda Aqa Soltan, the number of tweets in PE category is always larger than the other two categories. This means most of the tweets were written in English by Persian speaking users, perhaps in order to report the events to the outside world.
Peaks of the graph match the most important events of those days, to name a few, when Neda Aqa Soltan was killed, and the days when protests and marches were held e.g. Qods, Student's, and Scholar's days. Most of the peaks occurred at the same time in all three sets of tweets.
Fig. 10. Number of tweets sent each day
Fig. 11. Total number of tweets mentioning famous politicians' name
4.4. Politician's names
Number of tweets referring 4 famous Iranian politicians in the post-election events, i.e. Khamenei, Ahmadinejad, Mousavi and Karroubi, has been calculated. Figure 11 gives the number of tweets in the three categories, separately and totally. Mousavi, Khamenei and Ahmadinejad are referred the most in all categories. Hashemi is mostly referred in English tweets (E and PE) however Karroubi is more known in P tweets and has less international fame. Khamenei is more referred in P Tweets than Ahmadinejad while, Ahmadinejad is more referred in E Tweets; both are referred almost equally in PE tweets.
Figure 12 shows the number of tweets sent per day in each category, normalized by the total number of tweets in the corresponding category. Some of the important events have been marked on the figure. The peaks match the day of events or sometimes they continue one or more days after the event. If a peak had happened before an event, we could imagine that Twitter have had been used for organization of that event, but we did not came across such events in the dataset. This means the events have only been lively reported and the outside world has not been directly influential on the events. Twitter has been used as a media to report the events to outside of Iran in order to break the monopoly of government over the media, inform abroad, and ask for sympathy and support.
5. Content analysis of tweets
In this paper, we analyzed the tweets that had been sent about Iran election in Twitter by Persian and English speaking users. We showed that Iranian users started to actively use Twitterjust a few days after the election when some mobile services were taken down by the telecommunication companies. This is the time when most active users were joined Twitter and formed a social network with scale free structure. The same argument has been made in a similar study by (Zhou et al, 2010); however their study mainly focuses on the structure and diffusion of tweets while our study is focused on the contents of tweets and mutual effects of Twitter as a virtual environment with real world.
We showed that the number of tweets per day could be used as a mean to find whether something important (i.e. a trend) has occurred. We also showed that tweets were mainly used to communicate the events to the outside world. This means that Twitter was mainly used as a one-way media and had no significant direct effect on the events inside Iran. Therefore importance of Twitter in the post-election events should not be over-estimated.
We also showed that users interested in post-election events could be divided into two almost separate communities, based on the age of membership in Twitter; old users were mainly communicating with old
We analyzed the content of tweets and found the most frequent words and patterns. We showed that politician names were mentioned in their related events. So appearance of a name in the tweets could mean something important around him had happened. We saw that some politicians were influential in the events inside Iran while being unknown to outside world.
For future we would like to study the diffusion of tweets and the relation between content and successful diffusion.
Fig. 12. Number of tweets per day mentioning famous politicians' name
Acknowledgements
This research was in part supported by a grant from IPM. (No. CS1390-4-06).
References
Adamic, L., & Glance, N. (2005). The Political Blogosphere and the 2004 U.S. Election: Divided They Blog. World Wide Web 2005, Conference's2ndAnnual.
AleAhmad, A., Amiri, H., Darrudi, E., Rahgozar, M., & Oroumchian, F. (2009). A standard Persian text collection. Journal of Knowledge-Based Systems, 382-287.
Barabasi, A., & Albert, R. (1999). Emergence of scaling in random networks. Science, 509-512.
Beil, F., Ester, M.,& Xu, X. (2009). Frequent term-based text clustering. Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining.
Cottle, S. (2011). Media and the Arab uprisings of 2011: Research notes. Journalism, 647-659.
Elsson, J., Gansner, E., North, S., & Woodhull, G. (2003). Graphviz and dynagraph - static and dynamic graph drawing tools. In GRAPH DRAWING SOFTWARE (pp. 127-148). Springer-Verlag.
Etling, B., Kelly, J., Faris, R., & Palfrey, A. (2009). Mapping the Arabic Blogosphere: Politics, Culture, and Dissent. Berkman Center Research Publication No. 2009-06.
Fruchterman, T., & Reingold, E. (1999). Graph drawing by force-directed placement. Software: Practice and Experience, 11291164.
Ghannam, J. (2011). Social Media in the Arab World: Leading up to the Uprisings of 2011. Washington D.C.: Center for International Media Assistance.
Hannon, J., Bennett, M., & Smyth, B. (2010). Recommending twitter users to follow using content and. proceedings of the 4th ACM Conference on.
Kelly, J., & Etling, B. (2008). Mapping Iran's Online Public: Politics and Culture in the Persian Blogosphere. Berkman Center Research Publication No. 2008-01.
Khonsari, K., Nayeri, Z., Fathalian, A., & Fathalian, L. (2010). Social Network Analysis of Iran's Green Movement Opposition Groups Using Twitter. Advances in Social Networks Analysis and Mining.
Lewis, D. (1999). Reuters-21578 text categorization test collection Distribution 1.0. available online at: http://www. research.att.com/~lewis.
Owen, P., Kevin, M., & Barry, S. (2009). Using Twitter to recommend real-time topical news. Proceedings ofthe third ACM conference on Recommender.
Qazvinian, V., Rassolian, A., Shafei, M., & Adibi, J. (2007). A Large-Scale Study on Persian Weblogs. Workshop ofTextLink.
Rane, H., & Salem, S. (2012). Social media, social movements and the diffusion of ideas in the Arab uprisings. Journal of International Communication, 97-111.
Zhou, Z., Bandari, R., Kong, J., Qian, H., & Roychowdhury, V. (2010). Information resonance on Twitter: watching Iran. Proceedings of the First Workshop on Social Media Analytics,ACM, (pp. 123-131).
