On the Way to a Science Intelligence:
Visualizing TEL Tweets for Trend Detection
Peter Kraker1, Claudia Wagner2, Fleur Jeanquartier1, and Stefanie Lindstaedt1
1Know-Center, Ineldgasse 21a, 8010 Graz, Austria
{pkraker,fjean,slind}@know-center.at
2Joanneum Research, Steyrergasse 17, 8010 Graz, Austria
claudia.wagner@joanneum.at
Abstract. This paper presents an adaptable system for detecting trends
based on the micro-blogging service Twitter, and sets out to explore
to what extent such a tool can support researchers. Twitter has high
uptake in the scientic community, but there is a need for a means of
extracting the most important topics from a Twitter stream. There are
too many tweets to read them all, and there is no organized way of
keeping up with the backlog. Following the cues of visual analytics, we
use visualizations to show both the temporal evolution of topics, and
the relations between dierent topics. The Twitter Trend Detection was
evaluated in the domain of Technology Enhanced Learning (TEL). The
evaluation results indicate that our prototype supports trend detection
but reveals the need for rened preprocessing, and further zooming and
ltering facilities.
Keywords: science 2.0, trend detection, social media, qualitative anal-
ysis
1 Introduction
Twitter has high uptake in the scientic community. According to a recently pub-
lished survey, personal email, Twitter, Skype, and project mailing lists are the
most popular applications used for disseminating information by Semantic Web
researchers [14]. The main motivations for publishing and sharing content on
Twitter named by survey participants were: (1) to share knowledge about their
eld of expertise, (2) to communicate research results, and (3) to expand their
network. The fact that the two main reasons for researchers to use microblogging
services are communicating their research results and sharing information about
their eld of expertise makes Twitter a rich source of information, which can be
exploited to detect research trends. It seems to be a reasonable assumption that
the results of this study can be transformed to other technology-rich research
elds such as Technology Enhanced Learning.
This paper presents an adaptable system for detecting trends based on the
micro-blogging service Twitter, and sets out to explore to what extent such a
tool can support researchers. In the context of this work we dene a "trend" as

2 Peter Kraker, Claudia Wagner, Fleur Jeanquartier and Stefanie Lindstaedt
a term belonging to a topic which gains considerable interest during a certain
period of time. Similar to the TF/IDF measure from Information Retrieval, the
interestingness of an item can be dened as the number of occurrences of that
item in time interval i out of a larger interval j [6].
Our research revealed that there is a need to have a means of extracting the
most important topics from a Twitter stream. According to our evaluation (see
Section 3), there are too many tweets to read them all, and there is no organized
way of keeping up with the backlog. Finding something interesting is more of
a coincidence than the result of a structured search, even with tools that allow
for various lists of users and hashtags. What makes it even worse is the large
amount of noise generated by super
uous postings. Twitter's trending topics do
not help with that as they are not related to research.
Following the cues of visual analytics, we use visualizations to show either
the temporal evolution of topics, or the relations between dierent topics. We
developed a focused Twitter Crawler which uses the Twitter Streaming API [20].
The crawler can be adapted to any domain, by either (a) specifying a taxonomy
of keywords, (b) specifying a list of users, or (c) a combination of both. On
the client-side, our system provides two explorative visualization components:
a streamgraph for analyzing trends over time and a co-occurrence network for
analyzing semantic networks of terms which may for example reveal which topics
are popular at the moment or which topics are strongly correlated.
The Twitter Trend Detection was evaluated in the domain of Technology
Enhanced Learning (TEL). The system was adapted to the domain using a
taxonomy of 30 hashtags, and a list of 450 expert users from TEL. First, we
conducted semi-structured interviews involving the use of the system with re-
searchers from the domains of TEL and knowledge management. The outcomes
from these interviews were used to further improve the system. It was then eval-
uated a second time at the 2nd STELLARnet Alpine Rendez-Vous, where the
visualizations were used as a means of support. The evaluation results indicate
that our prototype supports trend detection but reveals the need for rened
preprocessing, and further zooming and ltering facilities.
1.1 Related Work
Past research, such as ThemeRiver [9], was exploring solutions for automatically
detecting emerging trends from collections of documents. With the rise of user
generated content, researchers started exploring to what extent social media
can be used to detect and monitor emerging trends. Fukuhara [7], for example,
presents a system which generates a daily trend graph of weblog articles con-
taining any given keyword. Glance et al. [8] introduce a tool called BlogPulse
which allows monitoring trends in weblogs. They show a correlation between
"blog and real world temporal data" such as temperature and news articles.
Hotho [11] presents an approach for discovering topic-specic trends within folk-
sonomies, by adapting PageRank algorithm to the triadic hypeWeighted Graph
structure of a folksonomy.

On the Way to a Science Intelligence 3
Cheong et al. [5] describe an approach of analyzing trend patterns on Twitter.
They explore the properties and features of a trending topic and the properties
of the users (or 'trend setters') that contribute 'tweets' to a trending topic,
which makes them a part of the trend. For analyzing trending topics versus
non-trending topics, they used only the last 1500 tweets which one obtains from
the Twitter search API when conducting a keyword search. Mathioudakis [15]
introduces TwitterMonitor, a system that discovers "bursty" keywords1 in Twit-
ter streams and uses them as 'entry points' for trend detection. Finally, Ben-
hardus [2] also presents a trend detection system based on Twitter. He uses term
frequency-inverse document frequency analysis and relative normalized term fre-
quency analysis to identify the trending topics.
Our work diers from previous work by focusing on supporting researchers
during their daily work with domain-specic information. Most existing work
deals with general trends on Twitter which are largely irrelevant to researchers.
To the best of our knowledge, this is the rst system that can be adapted to a
certain domain. The prototype's UI is a web-based one that relies on web stan-
dards. Therefore the visualization can be displayed in a standard web browser,
but also be easily integrated with any system that allows for widgets adhering
to the W3C standard. This design is far more adjustable compared to existing
solutions.
2 System
2.1 Twitter Stream Analysis
We developed a focused Twitter Crawler which uses the Twitter Streaming API
[20]. The crawler can be adapted to any domain, by either (a) specifying a tax-
onomy of keywords, (b) specifying a list of users, or (c) a combination of both.
The Twitter Crawler then logs only tweets which were either authored by a
certain user or which contain at least one keyword of the taxonomy. Next, we
lexically preprocess the logged tweets in order to extract "informative" tokens
(mainly nouns and hashtags) from it, using a part-of-speech tagger (POS Tag-
ger), namely TreeTagger [16]. Finally we store the tweets, their metadata, and
their associated informative tokens in a Solr index [18] which is used by our
Visualization Dataservices. Fig. 1 illustrates this architecture.
The Visualization Dataservices are REST-ful Webservices which enable cli-
ent-side, lightweight visualizations to fetch and reload the data on demand.
Since we provide two dierent types of visualizations, streamgraphs and weighted
graphs, we also implemented two dierent types of Visualization Dataservices.
Both Visualization Dataservices take the following parameters as input: (1)
a query consisting of one or several search terms (e.g., "conferences") (2) a time
period of interest (e.g., 10.4.2010 - 15.5.2010), (3) the maximum number of co-
occurring terms, and (4) the type of co-occurring terms (either nouns, hashtags,
or users). Besides, the Streamgraph Dataservice optionally takes the number of
1 Keywords which are frequently used within a short period of time.

4 Peter Kraker, Claudia Wagner, Fleur Jeanquartier and Stefanie Lindstaedt
time intervals as additional parameter. Both Dataservices translate the search
query into a Solr query and preprocess the Solr result in dierent ways: the
Streamgraph Dataservice focuses on analyzing the temporal evolution of topics
over time; the Weighted Graph Dataservice focuses on relations between dierent
topics.
The Streamgraph Dataservice splits the time-period of interest into the spec-
ied number of intervals. For each interval, the Streamgraph Dataservice returns
the most frequent topics which co-occur with the query term within said inter-
val. The Weighted Graph Dataservice returns the most important topics which
co-occur with the query terms and the number of times they co-occurred.
Fig. 1. Architecture

On the Way to a Science Intelligence 5
2.2 Visualizations
Following the cues of visual analytics [12], we use visualizations to show both the
temporal evolution of topics, and the relations between dierent topics. First,
we studied existing visualization techniques to identify those that are suitable
for our purpose.
Heer et al. [10] describe a stacked graph as a classic method for visualizing
change in a set of items, where the sum of the values is as important as the
individual items. While such charts have proven popular in recent years, they
do have some limitations such as the fact, that stacking may make it dicult
to accurately interpret trends that lie atop other curves. The authors in [9]
describe a more enhanced visualization technique that is suitable for visually
describing thematic variations over time within a large collection of documents.
This technique makes use of a river metaphor for revealing patterns and trends.
One of the major advantages of this technique is little dependence on the number
of documents; furthermore, the stacked areas are suitable for observation and
comparisons. Inspired from the visualization technique mentioned before, Byron
and Wattenberg further describe the Streamgraph design, a unied approach to
stacked graph geometry and algorithms [4]. We therefore identied this special
stacked graph design technique called "streamgraph" as one suitable approach
to visualize Twitter-data over time. Lamping et al. [13] describe a focus+context
visualization technique that is intended for displaying hierarchies, which is the
basis for our second visualization approach: visualizing the relations between
topics.
We further extended our visualization study to comparing existing web script
libraries for the two visualization techniques described above that could be used
in our prototype. The Streamgraph Visualization builds on top of the Graco
javascript library [22] while the Weighted Graph Visualization is based on the
Javascript InfoVis Toolkit [1]. An HTML-based frontend oers the possibility
to type in one or several search terms, furthermore specifying a facet such as
"hashtags that occurred with the term above". Furthermore, the UI provides a
date selection to narrow the search to a specic date range, and a eld to specify
the maximum number of co-occurring terms.
Fig. 2 shows a screenshot of the Streamgraph Visualization, displaying the co-
occurring hashtags for the query "conferences" from 20/2/2011 to 14/04/2011.
On the x-axis, the time intervals are outlined, whereas on the y-axis, the relative
number of occurrences is shown. Each colored stream represents one co-occurring
hashtag. The visualization shows that the hashtag for the South-by-Southwest
conference (#sxsw) is trending around the actual event on March 15.2 The
#pelc11 hashtag was trending around April 7, with the Plymotuh E-Learning
conference taking place from April 6-8. Another conference that is trending is the
PLE Conference in Southhampton which has not taken place yet (#PLE SOU).
The other co-occurring hashtags are not tied to a certain conference (such as
#mlearning and #edchat), but they denote hashtags in the TEL area which
2 The conference took place from March 13-20.

6 Peter Kraker, Claudia Wagner, Fleur Jeanquartier and Stefanie Lindstaedt
contain a large amount of tweets about conferences. These hashtags could there-
fore be used to nd out about other conferences in the area. An example for the
Weighted Graph Visualization is shown down below in section 3.2.
Fig. 2. Streamgraph Visualization
3 Evaluation
The Twitter Trend Detection was evaluated in the domain of Technology En-
hanced Learning (TEL). The system was adapted to the domain using a domain-
specic taxonomy of 30 hashtags. That removes the noise that is generated when
taking all tweets into account, or when using general keywords such as "learn-
ing" and "training". In an early prototype using general keywords, we found
that there is a strong correlation between "training" and "dog"; this is interest-
ing as a fact, but irrelevant to the domain of TEL. The taxonomy was created
(1) by analyzing hashtags that occur with the more general keywords, and (2)
by searching Twapperkeeper [19] for relevant archives. Besides the taxonomy, we
also created a list of 450 user accounts from various lists related to Technology
Enhanced Learning which belong to well-known domain experts, researchers, and

On the Way to a Science Intelligence 7
students. The system is part of the STELLAR Science 2.0 infrastructure [21].
The visualizations are also integrated as early-stage widgets on the TELeurope
platform [17].
The evaluation of the Twitter Trend Detection is based on the creative think-
ing method PMI [3]. PMI stands for Plus, Minus, and Interesting. Instead of
asking participants to take an objective stance in the evaluation, their attention
is guided to think separately about (a) the positive aspects of the system (Plus),
(b) the negative aspects of the system (Minus), and (c) the neutral but note-
worthy aspects of the system (Interesting). This terminology was not only used
in the instruments, but also later in the analysis, guiding the qualitative coding
scheme.
The Twitter Trend Detection was evaluated in two settings: rst, we held
semi-structured interviews involving the use of the system with ve researchers
from the domains of Technology Enhanced Learning and knowledge manage-
ment. The new system was then further evaluated at the 2nd STELLAR Alpine
Rendez-vous where the visualization was used as a means of support. We utilized
them in a re
ection session in one of the workshops; this was accompanied by
three in-depth interviews with conference participants.
3.1 Evaluation 1
In the rst evaluation, we held semi-structured interviews involving the use of the
system with ve researchers from the domains of Technology Enhanced Learn-
ing and knowledge management based in Austria. Among the participants were
one professor, three senior researchers, and one PhD student. Two had a back-
ground in computer science, two in psychology, and one in business administra-
tion. Participants were interviewed about their use of Twitter in research, and
specically in relation to trend detection. Afterwards, they were introduced to
the visualizations. Following a short tutorial, we asked participants to search for
trends in their area of interest. The interviews were recorded on tape, and later
transcribed. We qualitatively analyzed the transcripts using a reducing and in-
terpreting approach. Codes were divided into three sections, following the PMI
terminology: Plus, Minus, and Interesting. Each of these sections was subdivided
into the codes "General", "Weighted Graph", and "Streamgraph". To capture
the general remarks on Twitter, we added "Usage of Twitter", "Advantages
of Twitter", and "Disadvantages of Twitter" to the scheme. Initially, the tran-
scripts were coded with these general codes. In a second iteration, we rened
these codes to paraphrase the content of marked statements. In the last step, we
merged similar paraphrases to remove redundancy in the scheme.
The goal of the rst evaluation was to collect feedback on the usability and
the general applicability of the system to trend detection. In this phase, the
system contained three dierent visualizations: two competing streamgraph vi-
sualizations, and one Hypertree Visualization. Table 1 shows the top two recom-
mendations that resulted from this rst evaluation. The evaluation showed that
users were struggling with the interface. It proved to contain too many parame-
ters that were labeled with technical terms. Furthermore, users had a hard time

8 Peter Kraker, Claudia Wagner, Fleur Jeanquartier and Stefanie Lindstaedt
interpreting the co-occurring terms, and they were unclear about the underlying
data. As a result, the user interface was completely overhauled, and we started
to display the rst 100 analyzed tweets alongside the visualizations.
Table 1. Recommendations from rst evaluation
General:
Redesign the user interface to make it more accessible to the user
Include more metadata for the co-occuring terms
Streamgraph:
Keep Graco streamgraph as it is more clearly laid out and has a better usability
Highlight meaning of the axis to the user
Hypertree:
Replace visualization with a version that is more clearly laid out
Position search term more prominently
The initial Hypertree Visualization was replaced by the Weighted Graph
Visualization, and from the two streamgraphs, only the Graco Streamgraph
was developed further. Axis descriptions were added to the Graco Streamgraph,
and in the Weighted Graph Visualization the search term was highlighted.
3.2 Evaluation 2
The new system was further evaluated at the 2nd STELLAR Alpine Rendez-
vous, where the visualizations were used as a means of support. Fig. 3 shows
the Weighted Graph of hashtags for the main conference hashtag "arv11" from
27/03/2011 to 14/04/2011. This covers the conference which took place from
28/03 to 01/04 as well as the discussion afterwards. The size of the circles in-
dicates the number of occurrences, while the line thickness indicates, how often
two hashtags co-occurred. The visualization is centered on "arv11". The hash-
tags on the rst level are directly related to arv11, namely "ngtel", "arvmupe-
mure", "multivocalanalysis", "datatel11", and "arv3t". They all represent dier-
ent workshops that were held during the conference. "jtelws11" represents the
JTEL Winter School which was co-located with the Alpine Rendez-vous. For
each of the workshops, as well as the winter school, a number of co-occurring
hashtags are identied on the second level that tell a bit more about the indi-
vidual workshops.
In the dataTEL workshop, we presented the corresponding visualizations in
a re
ection session on the workshop. During the course of the presentation and
the ensuing discussion, participants were asked to list positive (Plus), negative
(Minus), and neutral but noteworthy (Interesting) aspects on post-its. This eval-
uation was complemented with three in-depth interviews with participants from
the Alpine Rendez-vous involving the visualizations. The interviewees were all
tweeting in the course of the conference, and were specically asked about the
usefulness of the system. Among them were one professor, one senior researcher,

On the Way to a Science Intelligence 9
Fig. 3. Weighted Graph Visualization
and one PhD student. Two had a background in education, and one in computer
science. Participants came from Europe, the United States, and Canada. The re-
sults were recorded, transcribed, and analyzed in the same manner as described
in the rst part of the evaluation above.
The evaluation showed that Twitter is regarded as an important means of
communication among the participating TEL researchers. Interviewees found it
interesting (a) to follow and to contribute to the backchannel discussion in their
own workshop, thus enriching their experience, and (b) to follow what is going on
in other workshop. They also used Twitter to document parts of the workshop,
and to keep their teams at home up-to-date, sometimes even using designated
hashtags for that.3 Among the uses outside of conferences were (1) to use it as
a source of information, (2) to ask for feedback on one's own work, and (3) to
directly communicate with other researchers.
What had already surfaced in the rst evaluation, was also repeatedly noted
in the second evaluation: there is a need to have a means of extracting the most
important topics in a Twitter stream. According to the participants, there are too
many tweets to read them all, and there is no organized way of keeping up with
3 One of these hashtags "#yam", can be seen in Fig. 3 as a co-occurring hashtag of
"#dataTEL11".

10 Peter Kraker, Claudia Wagner, Fleur Jeanquartier and Stefanie Lindstaedt
the backlog. As one of the interviewees put it so aptly, "If I get up to get coee,
I could have already missed something important." For the interviewees, nding
something interesting is more of a coincidence than the result of a structured
search, even with tools that allow for various lists of users and hashtags. What
makes it even worse in the eyes of the participants is the large amount of noise
generated by super
uous postings ("I am having breakfast now"). Twitter's
trending topics do not help with that as they are not related to research.
3.3 Discussion
Participants liked the looks of the visualizations, and the idea behind them. In
both evaluations, they noted that the interface is visually appealing. They also
noted on several occasions that the system might be a useful way to deal with the
backlog in their Twitter streams. The two visualizations are complementing each
other very well; participants were interested in the connections between topics as
well as the temporal evolution. Participants noted that both of them condense
a lot of information in one view. They enjoyed the fact that the visualizations
operate on live data with the ability to go back in time. Another feature that
was well received was the consistency between the two visualizations, as the
visualizations always operate on the same set of tweets for a given search term
and a given time range.
As for the Weighted Graph, people were easily able to understand the basic
visual metaphor, albeit it's sometimes crowded nature, and the fact that the
weighted Graph is not always centered on the initial search term. Most inter-
viewees could instantly interpret the size of the nodes and the thickness of the
edges correctly. It was only the dierent levels that were hard to grasp in some
cases. Participants noted that edges between topics are useful to determine the
connection between the two, and the kind of clustering that is provided by the
Weighted Graph in that way. They recognized topics from the discussions in
their workshops, as well as users which participated in the backchannel discus-
sion. An additional use that they saw was to get a quick overview of a eld that
they were not familiar with.
In the case of the Streamgraph, participants that were not accustomed to the
visual metaphor needed a bit more time to understand the concept. Especially
the alignment and the color-coding were often mis-interpreted. After a short
introduction though, they liked that topics are shown in such a way that one
can clearly see the bursty terms. They were able to reconstruct the time wise
evolution of certain discussions from their workshops. One participant said it
would be interesting to have such a visualization running on several screens
at a conference to keep everyone updated about the current sessions. Another
possible use that was mentioned is the detection of pivotal moments in online
discussions.
Despite all the positive feedback, the evaluation also pointed out several
shortcomings of the visualizations. First and foremost, users would like to be
able to zoom into the results. They would like to be able to click on a co-
occurring term to see its metadata, and they want to be able to not only see a

On the Way to a Science Intelligence 11
full list of tweets but rather a ltered one. This revealed the need to implement
zooming and ltering and the need to provide more details on demand. This
should be complemented by a short help page on either visualization to make
the metaphor crystal clear. In addition, users demanded more meaningful terms.
They found the co-occurring terms to be too broad and generic. In addition, we
need to get better at ltering out hashtags and mentions, and ignoring the case
in the output. This revealed the need for rened preprocessing.
On a meta-level, participants sometimes criticized Twitter as a data base, as
they trust only certain experts with trends. To address those concerns, we need
to nd a way to include only certain user accounts in the search. Finally, users
would like to be able to integrate the visualizations with their existing Twitter
applications.
4 Outlook
The evaluation results indicate that our prototype supports trend detection, but
we still need to address several issues. First and foremost, we will provide more
insight into the data in connection with further ltering mechanisms to allow
users to view only a portion of that data. We will also improve the data quality
by changing our crawler so that it does not count users and hashtags as nouns,
and by applying lowercase to all output terms. To weed out the generic terms,
we will look into applying the TF-IDF measure, and/or blacklists of common
and broad terms.
Moreover, we will look into ways of integrating the visualizations with ex-
isting platforms emerged. We already took a rst step into that direction by
creating a W3C compliant widget, which can be included into any system that
allows for such widgets; a rst version has already been deployed to the social
network TELeurope. Furthermore, it would be interesting to have a kind of self-
evolving taxonomy of hashtags and users which semi-automatically adds new
pieces of information to the taxonomy. On the front-end side, we are constantly
looking into new meaningful visualizations, which can be adapted for system.
With the ever increasing amount of tweets, scalability becomes an issue; in
a little over three months, we have collected over 500.000 tweets. Nevertheless,
Solr has proven to be able to go way beyond that number of documents. The
system is currently tailored towards Technology Enhanced Learning, but it could
easily be adapted to other elds of research. The only precondition is to produce
a list of hashtags and/or users from the eld. If such a taxonomy does not exist,
the system itself can help by detecting co-occurrent hashtags starting even from
a single high-frequency hashtag.

12 Peter Kraker, Claudia Wagner, Fleur Jeanquartier and Stefanie Lindstaedt
Acknowledgements
We would like to thank all participants in the evaluation. This work was car-
ried out as part of the STELLAR Network of Excellence, which is funded by
the European Commission (grant agreement no. 231913). The Know-Center is
funded within the Austrian COMET program - Competence Centers for Ex-
cellent Technologies - under the auspices of the Austrian Federal Ministry of
Transport, Innovation and Technology, the Austrian Federal Ministry of Econ-
omy, Family and Youth, and the State of Styria. COMET is managed by the
Austrian Research Promotion Agency FFG. Claudia Wagner is a recipient of a
DOC-fForte fellowship of the Austrian Academy of Science.
References
1. Belmonte, N.G.: JavaScript InfoVis Toolkit, http://thejit.org/
2. Benhardus, J.: Streaming Trend Detection in Twitter. Tech. rep., University of
Colorado at Colorado Springs (2010)
3. Bono, E.D.: De Bono's thinking course. Pearson Education (2006)
4. Byron, L., Wattenberg, M.: Stacked graphs{geometry & aesthetics. IEEE transac-
tions on visualization and computer graphics 14(6), 1245{52 (Jan 2008)
5. Cheong, M., Lee, V.: Integrating web-based intelligence retrieval and decision-
making from the twitter trends knowledge base. In: Proceeding of the 2nd ACM
workshop on Social web search and mining - SWSM '09. pp. 1{8. ACM Press, New
York, New York, USA (2009)
6. Dubinko, M., Kumar, R., Magnani, J., Novak, J., Raghavan, P., Tomkins, A.:
Visualizing tags over time. ACM Transactions on the Web 1(2) (Aug 2007)
7. Fukuhara, T.: Analyzing concerns of people using Weblog articles and real world
temporal data. In: Proceedings of the 14th International Conference on World
Wide Web - WWW 2005 (2005)
8. Glance, N., Hurst, M., Tomokiyo, T.: Blogpulse: Automated trend discovery for
weblogs. In: WWW 2004 Workshop on the Weblogging Ecosystem: Aggregation,
Analysis and Dynamics. vol. 2004. Citeseer (2004)
9. Havre, S., Hetzler, E., Whitney, P., Nowell, L.: ThemeRiver: visualizing thematic
changes in large document collections. IEEE Transactions on Visualization and
Computer Graphics 8(1), 9{20 (2002)
10. Heer, J., Bostock, M., Ogievetsky, V.: A tour through the visualization zoo. Com-
munications of the ACM 53(6), 59 (Jun 2010)
11. Hotho, A., Jaschke, R., Schmitz, C., Stumme, G.: Trend detection in folksonomies.
In: Avrithis, Y.S., Kompatsiaris, Y., Staab, S., O'Connor, N.E. (eds.) First In-
ternational Conference on Semantics And Digital Media Technology (SAMT). pp.
56{70. Springer (2006)
12. Keim, D., Kohlhammer, J., Santucci, G., Mansmann, F., Wanner, F., Schafer, M.:
Visual Analytics Challenges. In: eChallenges 2009. Istanbul, Turkey (2009)
13. Lamping, J., Rao, R., Pirolli, P.: A focus+context technique based on hyperbolic
geometry for visualizing large hierarchies. In: Proceedings of the SIGCHI Confer-
ence on Human Factors in Computing Systems - CHI '95. pp. 401{408. ACM Press,
New York, New York, USA (May 1995)
14. Letierce, J., Passant, A., Breslin, J., Decker, S.: Understanding how Twitter is used
to spread scientic messages. In: Web Science Conference. Raleigh, NC (2010)

On the Way to a Science Intelligence 13
15. Mathioudakis, M.: TwitterMonitor: Trend Detection over the Twitter Stream. In:
Proceedings of the 2010 International Conference on Management of Data. pp.
1155{1157 (2010)
16. Schmid, H.: Probabilistic part-of-speech tagging using decision trees. In: Proceed-
ings of the International Conference on New Methods in Language Processing.
Manchester, UK (1994)
17. STELLAR Network of Excellence: TELeurope.eu, http://www.teleurope.eu/
18. The Apache Software Foundation: Apache Solr, http://lucene.apache.org/
solr/
19. Twapperkeeper.com: Twapper Keeper, http://twapperkeeper.com/
20. Twitter Inc.: Twitter Streaming API, http://dev.twitter.com/pages/
streaming_api
21. Ullmann, T.D., Wild, F., Scott, P., Duval, E., Parra, G., Reinhardt, W., Heinze,
N., Kraker, P., Fessl, A., Lindstaedt, S., Nagel, T., Gillet, D.: A Science 2.0 In-
frastructure for Technology-Enhanced Learning. In: Wolpers, M., Kirschner, P.,
Scheel, M., Lindstaedt, S., Dimitrova, V. (eds.) Sustaining TEL: From Innovation
to Practice. Proceedings of the 5th Conference on Technology Enhanced Learning.
pp. 590{595. Springer, Barcelona, Spain (2010)
22. Valkhof, K.: Graco javascript charting library, http://grafico.kilianvalkhof.
com/