A Tool for Monitoring of YouTube Content
Intzar Ali Lashari and Uffe Kock Wiil
The Maersk Mc-Kinney Moller Institute, University of Southern Denmark
Campusvej 55, DK-5230 Odense M, Denmark
Keywords:
Social Media, YouTube, Monitoring, Information Visualization, Social Network Analysis and Mining.
Abstract:
The expansion in use of social media has been very significant in the past decade. It has become a topic of
interest for many researchers to find the most connected people, the most influential people, etc. in social
media for various purposes. However, collection and monitoring of data in abundance from social media
is difficult. This paper describes a new tool that can collect, monitor, and mine data from YouTube. The
tool is part of a larger framework aimed at monitoring various social media including Facebook, Twitter, and
YouTube. A specific case focusing on “Islamic Jihad Holy War” demonstrates the features of the tool.
1 INTRODUCTION
A rapidly increasing number of people share informa-
tion with others by using different social media sites
such as Facebook, Twitter, YouTube, LinkedIn, MyS-
pace, etc. A social media site provides a platform to
the users for the social relations based on their respec-
tive interests, relations, and activities which they want
to share with their social circle on the internet. Social
networks can be described as information networks in
which actors are represented by nodes and relations
are represented by edges (Aggarwal, 2011). The con-
text of a social network works as a motivation for the
actors in the social network and the resulting content
generated by the actors and the structure encourages
the participation and subsequently affects the social
relation (Zeng and Wei, 2013). The increase in use
of social media has created an interest in the research
community to analyse and mine the social media data.
However, social network analysis and mining is be-
coming more and more challenging as the generated
content becomes richer and more abundant (Aggar-
wal, 2011).
Social media is used for many different purposes.
Three examples are given here: (1) In countries
like Egypt, Tunisia, and Yemen, rising action plans
such as protests made up of thousands, have been
organized through social media such as Facebook,
YouTube, and Twitter. They used Facebook to sched-
ule the protests, Twitter to coordinate, and YouTube
to tell the world as part of the Arab Spring upris-
ings (2012) (Polymic, 2012). (2) Social media was
extensively used during the East Japan Earthquake
(2011) to share information about the disaster and get-
ting in touch with missing relatives (Telegraph, 2011).
(3) Social media is used increasingly for militant is-
lamist propaganda with the intent to radicalise Mus-
lims (CTA, 2012). Hence, their is an increasing in-
terest in the ability to monitor the social media for
instance to get information about evolving events and
in the interest of public safety.
In this paper, we present a tool for monitoring of
YouTube content. The tool is part of a larger frame-
work aimed at monitoring social media such as Face-
book, Twitter, and YouTube. On YouTube various
types of videos can be uploaded by different users.
Users can watch, like/dislike, and comment upon the
available videos. It is possible to see how many times
a video has been viewed as well as other metrics about
videos and comments. Data changes over time as peo-
ple engage actively on YouTube. Section 2 describes
YouTube in more detail. A set of monitoring metrics
are proposed that can help find the most influencial
people, videos, etc. on YouTube based on social net-
work analysis and mining. Section 3 describes related
work. Section 4 presents the tool and Section 5 fo-
cuses on a specific case that is used to demonstrate
the features of the tool. Finally, Section 6 concludes
the paper.
2 YOUTUBE
What is the nature of user activity on YouTube? It is
171
Ali Lashari I. and Kock Wiil U..
A Tool for Monitoring of YouTube Content.
DOI: 10.5220/0005068401710178
In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2014), pages 171-178
ISBN: 978-989-758-050-5
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
an assumption by some scholars (Tannen, 1999) that
in on-line environments the prevalence of anonymity
directly spawns antagonism, and that an increase in
identity information will decrease the communica-
tive hostility. However some scholars hold the op-
posing view that additional identity information, such
as, facial and bodily information, does not guaran-
tee cordial communication among the commentators
(Lange, 2007).
On-line communicative environments like video
sharing sites, blogs, etc. provide a platform for ex-
change of comments. The comments can have a
wide range of reaction, ranging from ecstatic praise
to extreme hate to threat of physical violence. How-
ever, given the fact that on-line interactions take place
in relative anonymity, not all participants may view
certain critical comments as a problem that requires
greater regulatory control, which can be viewed as a
threat to limit participation (Lange, 2007).
Two major types of negative comments can be
distinguished—comments that are hateful or threaten
violence and comments that provide constructive crit-
icism. Hateful comments are often the cause of dis-
couragement of open self-expression on the site. The
driving component in such on-line exchange of com-
ments are the users of video exchange site, blogs, etc.
who participate in the textual communication. Ac-
cording to a previous study, the participants can be
classified in one of several categories (Lange, 2007):
1. Former Participants. They are those who no
longer post videos, blog posts, etc. but still main-
tain an account.
2. Casual Users. They typically don’t have an as-
sociated account. However, they tend to view
videos or read blogs, etc. when they wish to
search for something while surfing the Web, and
are prompted to view a video or read a blog post.
3. Active Participants. They usually have an as-
sociated account, and occasionally participate by
uploading videos, writing posts, and/or by com-
menting on other people’s contributions.
4. Highly-active Participants. These have a more
intense level of activity and participation on the
social sites, spend much more time regularly up-
loading content and maintaining their sites. They
tend to promote their work within and outside of
the content platform.
5. Celebrities. Similar in many respects to the last
category, but are also well-known despite their on-
line presence in the form of YouTube channels,
blogs, FaceBook pages, or Twitter handlers. They
are often in a position to influence discourse by
the content they upload/create, and other interac-
tions on such sites.
The above categories only provide a description
of the relative levels of participation among the whole
body of users and are not mutually exclusive.
In order to monitor user activity on YouTube, it is
necessary to think in terms of metrics that can quan-
tify users and their activities. Hence, important ques-
tions are: What would we like to monitor? And what
can be monitored given the available YouTube API
(YouTube, 2014)? In terms of videos, it is interesting
to see how influential they are. This can be measured
by the number of views, the number of likes/dislikes,
and the number of comments. In terms of users, it
is interesting to see how influential they are and who
they engage with. The former can be determined by
the number of videos they post and how influential
they are (see above). The latter can be found by ana-
lyzing the social network formed by the activities of
users based on who comments on what videos.
For the purpose of this work, we focus on videos
that are retrieved based on keyword-matching using
YouTube’s API. For each matching video, data about
users who have made comments on that video are col-
lected. The user data collected includes the YouTube
identifiers for users who uploaded the video and users
who subsequently commented on the video. This
enables us to generate a social network of YouTube
users from a contextual perspective, where the con-
text is defined by keywords typed in by the person
(investigator or analyst) that wishes to monitor a cer-
tain event, activity, etc.
3 RELATED WORK
Internet-based applications that are build on the tech-
nological and ideological foundation of Web 2.0 al-
lows users to create and exchange the contents of their
interest (Kaplan and Haenlein, 2010). These appli-
cations include Facebook, Twitter, and YouTube to
name a few.
It has been suggested that videos are a very potent
medium for affecting the attitudes and political will
of the intended audience (Farwell, 2010). They can
be used to communicate a message for influencing
values, culture, attitudes, and opinions. Online video
platforms, like YouTube, provide a very effective and
cheap way to reach mass audiences that would other-
wise be difficult to reach using conventional means.
Terrorist networks has received much attention af-
ter 9/11 (2001). The Al-Qaeda network and related Ji-
hadist organizations have been analyzed with respect
to their Internet based information strategies. Related
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
172
work in this area has considered online content of Ji-
hadists and their supporters (Conway, 2006) with a
particular emphasis on the content of Jihadi videos
including various types of videos and their impact on
their audiences (Salem et al., 2006), (Kimmage and
Ridolfo, 2007), (Salem et al., 2008).
For example, the HBO documentary called Bagh-
dad ER (Baghdad, 2006), which dealt with the sub-
ject of providing emergency medical care to wounded
US personnel in the battlefield, was re-created by Al-
Qaeda based in Mesopotamia by replacing the origi-
nal soundtrack with their own, and by making an en-
tirely different beginning and ending to show that the
US forces are sustaining loses and being defeated in
combat.
Similarly, in Iraq and Afghanistan, terror groups
use videos to demonstrate their victories over the op-
posing side. Viewer-enthusiasm is gauged by how
quickly the video spreads over the internet and into
news media sites (Farwell, 2010). Because of this un-
conventional approach, they have a greater chance of
getting news coverage via satellite TV which has a
large number of viewers in the Arab world and simi-
lar conflict zones.
The main idea is to identify the impact of videos
to the audiences. Funders and policy makers have
shown an increased interest in learning the ways of vi-
olent radicalization (Council of the European Union,
2005). A high profile example is Hussain Osman, one
of the London bombers, who claimed to have been
influenced by watching videos of the conflict in Iraq
along with reading about jihad in an online forum.
In this paper, we create a network of YouTube
video uploaders to investigate the possibilities of rad-
icalization via the Internet specifically from YouTube
as opposed to analysis of jihadist sites. We aim to find
the users that are registered on YouTube and have up-
loaded videos and/or have commented on video con-
tent related to a given context. Hence, our work is
closely related to the work of (Chen et al., 2003; Wen
et al., 2007; Das et al., 2008).
With respect to data collection, analysis, and vi-
sualization, Coplink (Chen et al., 2003) was one of
the first systems to successfully address the domain of
criminal network investigation. The system was first
deployed at the Tucson Police Department. The sys-
tem collects, combines, and analyzes data from vari-
ous sources and generates overviews of the informa-
tion for the investgators to help them solve cases.
With respect to the idea of context, (Wen et al.,
2007) presents an intelligent information system that
performs an investigation task for detecting frauds.
The authors have contributed by developing two no-
tions: 1. Context and 2. Context-awareness. Re-
lated to context, the paper defines the term investiga-
tion context and with regard to context-awareness the
investigator can adaptively retrieve data and evaluate
the relevant information for the ongoing investigation.
With respect to collection of user networks from
social media, (Das et al., 2008) focuses on improv-
ing performance in information collection of a so-
cial graph of users’ neighbours in a dynamic social
network. In the study, the author has introduced a
sampling based algorithm for quickly approximating
quantities of interest, the vicinity of a user’s social
graph that explores the variants of correlation across
the sample. The algorithm can be used to rank the
items in the neighbourhood of a user.
Pippal et al. (Pippal et al., 2014) provide a recent
survey of data mining approaches and methods in so-
cial networking sites, including micro-blogging, twit-
ter, YouTube, instagram, blogs, forums, etc. Closely
related work is done by Agarwal and Sureka (Ag-
garwal and Sureka, 2014) where the authors analyse
YouTube metadata for privacy invading and harrass-
ment content. He et al. model user comments on
YouTube videos as a bipartite graph to predict the
popularity of videos and other item on the Web. (He
et al., 2014)
The purpose of our study is to develop a tool for
context-based monitoring of social media data based
on a set of monitoring metrics. To our knowledge, no
one has defined monitoring metrics for YouTube with
the intent to develop a tool to explore and monitor
influential videos, users, and networks.
4 THE YOUTUBE MONITORING
TOOL
The YouTube monitoring tool is a part of our frame-
work entitled ”Keyword-based Social Network Anal-
ysis Framework” (KSNAF). KSNAF aims at support-
ing collection, monitoring, and mining of social me-
dia data from a contextual perspective. Overall, the
framework must meet the following requirements:
1. The framework enables context-based search. In
the case of YouTube, the framework provides fa-
cilities to collect videos that match a given context
based on keywords defined by the investigator.
2. The framework determines relationships among
the users that are retrieved as the result of a
context-based search. In the case of YouTube, the
framework can build a social network based on
the users’ activities (comments on videos).
3. The framework enables investigators to view
data related to a given set of monitoring met-
AToolforMonitoringofYouTubeContent
173
Figure 1: System architecture of KSNAF.
rics. In the case of YouTube, the framework can
point to the most influential people, videos, and
(sub)networks.
4. The framework supports social network analysis
related to the generated social network.
5. The framework can export data in formats that
are importable by various social network analysis,
mining, and visualization packages.
An architectural overview of the KSNAF frame-
work is shown in Figure 1. KSNAF includes the fol-
lowing components to meet the above described re-
quirements.
Comments and Authors Extractor. This com-
ponent is responsible for extracting commenters
and authors with respect to a particular video un-
der investigation.
Relationship Extractor. This component is re-
sponsible for extracting relationships as ordered
triplets of the form (Author, Commenter, Com-
ment).
Social Network Mapper. This component is re-
sponsible for mapping the relationships extracted
by the Relationship Extractor component into a
social network. This component contains the co-
ordinating logic that is used in interaction among
other components. At present, the component
uses Algorithm 1 to generate a context-aware so-
cial network.
Social Network Analyser. This component car-
ries out social network analysis on the networks
generated by the Social Network Mapper com-
ponent. It supports a number of network analy-
sis algorithms, computing various metrics for the
level of nodes, links, and the whole network. It
comes up with results such as what is the most fre-
quent Author-Commenter pattern with respect to
the searched context. Examples of network met-
rics computed by this component include mean
path length, density, and centrality measures (like
degree, betweenness, closeness, etc.). Finally, key
players, weighted link analysis (Memon, 2012)
and determining various clusters in the resulting
social network is supported by this component.
Analysis Reports. This component generates dif-
ferent reports on the basis of the Social Network
Analyser components’ findings over a particular
span of time.
Monitoring Social Network Evolution. This
component is responsible for determining the
changes over time in the structure obtained from
the user-generated activity of writing comments
and/or uploading videos. Hence any changes in
the network are identified and the network is up-
dated accordingly. These updates are done over
time by the Social Network Analyser component
in a particular search context.
Social Network Exporter. This component ex-
ports the social network in a variety of formats
to enable social network analysis, mining, and vi-
sualization of third-party software packages. At
present, the component supports comma sepa-
rated values (CSV), XML, and GraphML formats.
In Algorithm 1, V denotes the collection of all
videos, D gives the maximum allowable depth for
search, and d denotes the current search depth, A de-
notes the collection of authors who have commented
on the relevant videos, and finally G denotes the net-
work graph.
5 CASE STUDY AND RESULTS
In this section, we present a case study to demonstrate
the features of the developed YouTube monitoring
tool. The selected case is related to the uprising
(civil war) in Syria. According to the Danish Se-
curity and Intelligence Service as well as Danish
media (TV2, 2013), (DR, 2014), (Politiken, 2013),
around 90 Muslims have left Denmark to join
the ”holy war” in Syria. In particular, one video
on YouTube is heavily criticized for encouraging
Danish Muslims to join the civil war in Syria.
The video is entitled ”A Danish terrorist in Syria!”
(http://www.youtube.com/watch?v=VTJ1ynW60gU).
This clearly demonstrates the need for a monitoring
tool like the one described in this paper.
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
174
Data:
Videos V retrieved as result of searching with
context C on YouTube
D is the maximum allowed depth at for
searching
d is the current depth
Result:
Graph G of relationships on YouTube
for each video vx in V do
Retrieve the comment authors A;
Retrieve the comments C;
for each commenter cx in C do
Add an edge that connects A with cx in
G;
Retrieve the videos vx posted by Cx;
if d < D then
Add 1 to d;
Execute the Algorithm 1 with
arguments V x, D, d, G;
else
Return G
end
end
end
Algorithm 1: Searching the YouTube graph.
We have used the case study to validate the frame-
work. Basically, the tool is generic and any keyword-
based search of related videos and other associated
data, such as, user IDs, user locations, comments, etc.
can be retrieved from YouTube. The data which we
have collected from YouTube using the available API
(YouTube, 2014) is based on the context “Islamic Ji-
had Holy War.
We provide an analysis of the corresponding so-
cial network that was generated based on the data col-
lected from YouTube. Figure 2 depicts the results of
searching the context and yields the videos that are re-
lated to the context. The resulting videos are then fur-
ther analysed to determine what are the commenters
of the retrieved videos and what are other videos
posed by the commenter. Such analysis is beneficial
to determine the regular commenter on the videos in
the given context, to determine the key players and
sources that positively contributes in uploading the
most viewed or commented videos, and finally to re-
veal interesting communication patterns like upload-
comment-upload. Figure 3 shows the different social
networks behind each of the videos that is retrieved
against the context search. Table 1 shows the location
of the commenters.
We have run the tool using a Windows 7 PC hav-
ing the following configuration: Intel (R) Core (TM)
2 CPU6600 2.40GHZ and 4.00GB RAM. It took 6
Figure 2: Context-based search of “Islamic Jihad Holy
War”.
Table 1: Location of YouTube commenters.
Country of Users No. of Users
United States (US) 152
Great Britain (GB) 13
Germany (DE) 2
Sweden (SE) 6
Canada (CA) 16
Australia (AU) 6
minutes and 12 seconds to collect the data of author-
ship and generate the commenter clustering. It will
take more time for the collection of data if the depth of
the users’ interconnection is increased. For this case
we have used a depth of 5.
We have used the visualization tool Gephi and the
Yifan Hu graph drawing algorithm (Hu, 2005) which
is efficient and high quality in visualization of rela-
tionships. Figure 4 shows the users as nodes and the
relationships among them, as links, connecting users
who have commented on videos by other users. Al-
though, the networks show different snapshots col-
lected on different dates, the nodes in all the networks
have a one-to-one correspondence among themselves.
Figure 4b-f shows the evolution over time in the
network from March 20 to April 13. Figure 4a shows
the network on April 15. Hence, Figure 4 demon-
Figure 3: Multiple social networks extracted in the context
of the query.
AToolforMonitoringofYouTubeContent
175
(a) (b)
(c) (d)
(e) (f)
Figure 4: Network graphs for the extracted YouTube data. (a) shows the entire network (April 15, 2014), (b)–(f) networks
collected from March 20 to April 13, 2014 (newest first).
strates that we are able to monitor (track) how the in-
fluence of users and videos change over time. The
highly central users are those whose videos have gen-
erated most comments by other users (e.g., users 47,
92, 49, 86, 22, 91, etc.) as shown in Table 2. It
is interesting to observe that the highly central users
are connected among themselves, i.e., not only does
their own videos generate a large volume of user-
comments (they are popular and influential) but such
users are commenting on other popular videos as well.
Moreover, Table 3 shows the top most and least pop-
ular users in terms of the number of video views they
have received from other users in the network.
For the collection of data from YouTube we have
used three different API’s provided by YouTube: (1)
data collection based on keyword search, (2) data col-
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
176
Table 2: Comments on videos by highly central users.
User ID YouTube name No. of comments
47 Jman92854 734
92 meteosurreal 83
49 xHolyCrusader 64
86 Indago55 58
22 Phoebe Igor 55
91 wadeywilson101 48
Table 3: Users with highest and lowest view counts.
User ID Highest
view
count
User ID Lowest
view
count
111 158748 95 3
106 77002 175 12
116 50239 45 33
59 19310 174 35
120 19186 19 38
25 16631 27 38
88 7876 121 43
135 5501 169 44
177 5468 99 44
79 4861 12 46
lection on content authors, and (3) data collection on
author profiles. For the information collection of data
of the network we have connected the users informa-
tion on node ID which is the user’s YouTube ID and
related commenters data depending on the depth of
the relations on comments. There is not much varia-
tion in the location of the users from dataset. The ma-
jority of commenters belong to North America (US
and Canada). Figure 4b, the commenter with node ID
1 has claimed that “The Muslims dont need Crusaders
to embrace jihad, jihad is fundamental to Islam” This
is one of the examples of comments obtained from
the dataset based on the mentioned keywords. The
developed tool has the main purpose of supporting
user-defined keyword-based collection of data from
YouTube for monitoring.
6 CONCLUSION
The main goal of this study is to find YouTube users
who is uploading videos related to the chosen con-
text and those who have commented on those videos.
Our tool is able to extract data related to YouTube
videos based on keywords chosen by the investi-
gator. The collected data is stored in a relational
database for later operations and queries. The col-
lected data includes user attributes, user comments,
location, timestamps, etc. As far as our tool is con-
cerned, we are able to find relations of YouTube video
content and present them as a social network graph
that shows the relationships. In addition, various sta-
tistical metrics of the social network graph are pre-
sented. These metrics show the connectedness and
influence of users. Our tool is able to detect and up-
date the relationships on a regular basis and show that
on a specific time who is closely connected to whom
and who is interested in whose content and what con-
tent.
As part of the near term future work, additional
case studies will be made to further validate the
YouTube monitoring tool. Also, we wish to further
investigate how to best visualize the evolution of the
collected metrics and networks over time.
Monitoring tools for Facebook and Twitter are be-
ing developed in parallel to the described tool. They
will be documented in subsequent publications.
REFERENCES
Aggarwal, C. C. (2011). An introduction to social network
data analytics. Springer.
Aggarwal, N. and Sureka, A. (2014). Mining youtube meta-
data for detecting privacy invading harassment and
misdemeanour videos. Master’s thesis, Indraprastha
Institute of Information Technology, Dehli, India.
Baghdad (2006). Baghdad er. http://www.imdb.com/title/
tt0802944/.
Chen, H., Schroeder, J., Hauck, R. V., Ridgeway, L.,
Atabakhsh, H., Gupta, H., Boarman, C., Rasmussen,
K., and Clements, A. W. (2003). Coplink connect:
information and knowledge management for law en-
forcement. Decision Support Systems, 34(3):271–285.
Conway, M. (2006). Terrorism and the internet: New media
new threat? Parliamentary Affairs, 59(2):283–298.
Council of the European Union (2005). The european union
counter-terrorism strategy. Technical report, Council
of the European Union Report.
CTA (2012). Center for terror analysis danish se-
curity and intelligence service. militant islamist
propaganda on facebook and youtube (in dan-
ish). https://www.pet.dk/CTA//media/CTA/
UKLCTAanalyseFacebookogYoutubepdf.ashx.
Das, G., Koudas, N., Papagelis, M., and Puttaswamy, S.
(2008). Efficient sampling of information in social
networks. In Proceedings of the 2008 ACM workshop
on Search in social media, pages 67–74. ACM.
DR (2014). Danish television. 22 young muslisms from
aarhus has left for syria (in danish). http://www.dr.dk/
Nyheder/Ligetil/Dagens fokus/Indland/2014/01/22
unge muslimer fra Aarhus er rejst til Syrien.htm.
Farwell, J. P. (2010). Jihadi video in the war of ideas. Sur-
vival, 52(6):127–150.
AToolforMonitoringofYouTubeContent
177
He, X., Gao, M., Kan, M.-Y., Liu, Y., and Sugiyama, K.
(2014). Predicting the popularity of web 2.0 items
based on user comments. In Proceedings of the 37th
International ACM SIGIR Conference on Research
&#38; Development in Information Retrieval, SIGIR
’14, pages 233–242, New York, NY, USA. ACM.
Hu, Y. (2005). Efficient, high-quality force-directed graph
drawing. Mathematica Journal, 10(1):37–71.
Kaplan, A. M. and Haenlein, M. (2010). Users of the world,
unite! the challenges and opportunities of social me-
dia. Business horizons, 53(1):59–68.
Kimmage, D. and Ridolfo, K. (2007). Iraqi insurgent me-
dia: The war of images and ideas. Radio Free Eu-
rope/Radio Liberty Special Report, page 4.
Lange, P. G. (2007). Commenting on comments: Investigat-
ing responses to antagonism on youtube. In Annual
Conference of the Society for Applied Anthropology.
Retrieved August, volume 29, page 2007. Citeseer.
Memon, B. R. (2012). Identifying important nodes in
weighted covert networks using generalized centrality
measures. In 2012 European Intelligence and Security
Informatics Conference, pages 131–140. IEEE.
Pippal, S., Batra, L., Krishna, A., Gupta, H., and Arora,
K. (2014). Data mining in social networking sites:
A social media mining approach to generate effective
business strategies. International Journal of Innova-
tions & Advancement in Computer Science, 3(2).
Politiken (2013). Danish newspaper. danish muslisms are
encouraged to holy war in syria (in danish). http://
politiken.dk/indland/ECE2047926/danske-muslimer-
opfordres-til-hellig-krig-i-syrien/.
Polymic (2012). Twitter revolution: How the arab
spring was helped by social media. http://
www.policymic.com/articles/10642/twitter-revolution-
how-the-arab-spring-was-helped-by-social-media.
Salem, A., Reid, E., and Chen, H. (2006). Content analysis
of jihadi extremist groups videos. In Intelligence and
Security Informatics, pages 615–620. Springer.
Salem, A., Reid, E., and Chen, H. (2008). Multimedia con-
tent coding and analysis: Unraveling the content of
jihadi extremist groups’ videos. Studies in Conflict &
Terrorism, 31(7):605–626.
Tannen, D. (1999). The argument culture: Stopping Amer-
ica’s war of words. Ballantine Books New York.
Telegraph (2011). The telegraph. japan earthquake: how
twitter and facebook helped. http://www.telegraph.
co.uk/technology/twitter/8379101/Japan-earthquake-
how-Twitter-and-Facebook-helped.html.
TV2 (2013). Tv2 news (denmark). terrorism expert:
Danish jihad video is dangerous (in danish).
http://nyhederne.tv2.dk/article.php/id-70767751:
terrorekspert-dansk-jihadvideo-er-farlig.html.
Wen, Z., Zhou, M. X., and Aggarwal, V. (2007). Context-
aware, adaptive information retrieval for investigative
tasks. In Proceedings of the 12th international con-
ference on Intelligent user interfaces, pages 122–131.
ACM.
YouTube (2014). Youtube api v3. https://developers.
google.com/youtube/v3/.
Zeng, X. and Wei, L. (2013). Social ties and user content
generation: Evidence from flickr. Information Sys-
tems Research, 24(1):71–87.
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
178