Association and Temporality between News and Tweets

ania Moutinho

, Pavel Brazdil

1 a

and Jo

ao Cordeiro

1,2 b

LIAAD, INESC TEC – Institute for Systems and Computer Engineering, Technology and Science,

Rua Dr. Roberto Frias, 4200, Porto, Portugal

HULTIG – Centre of Human Language Technology and Bioinformatics, Universidade da Beira Interior,

Rua Marqu

es d’

Avila e Bolama, 6200, Covilh

a, Portugal

Keywords:

Text Mining, Temporal Analysis, Clustering of News, Evolution of Occurrence, Time-wise Differences.

Abstract:

With the advent of social media, the boundaries of mainstream journalism and social networks are becoming

blurred. User-generated content is increasing, and hence, journalists dedicate considerable time searching

platforms such as Facebook and Twitter to announce, spread, and monitor news and crowd check informa-

tion. Many studies have looked at social networks as news sources, but the relationship and interconnections

between this type of platform and news media have not been thoroughly investigated. In this work, we have

studied a series of news articles and examined a set of related comments on a social network during a period

of six months. Speciﬁcally, a sample of articles from generalist Portuguese news sources published on the

ﬁrst semester of 2016 was clustered, and the resulting clusters were then associated with tweets of Portuguese

users with the recourse to a similarity measure. Focusing on a subset of clusters, we have performed a tem-

poral analysis by examining the evolution of the two types of documents (articles and tweets) and the timing

of when they appeared. It appears that for some stories, namely Brexit and the European Football Cup, the

publishing of news articles intensiﬁes on key dates (event-oriented), while the discussion on social media is

more balanced throughout the months leading up to those events.

1 INTRODUCTION

The advent of social media is gradually changing the

way information is disseminated and, moreover, pos-

sibly shifting the roles of news makers and news re-

cipients. According to (Kaplan and Haenlein, 2010),

social media is the set of applications based on the

Internet where user generated content (UGC) is cre-

ated, modiﬁed and exchanged in a collaborative and

participatory way.

Journalism and social media have become more

intricately interconnected. Traditionally, people re-

sort to mainstream media to know what is happen-

ing in the world. However, this dynamic has been

changing in recent years, at least for some news top-

ics. This is due to the fact that a great proportion of

the world population has access to platforms which

broadcast real time events. According to data from

statista

(2019), 57.54% are active internet users, and

45.38% are active social media users. Consequently,

https://orcid.org/0000-0002-4720-0486

https://orcid.org/0000-0003-0466-1618

http://www.statista.com

the beginning of the process of news generation and

dissemination has in some cases changed the way

journalism is done. It has been recognized in vari-

ous studies (Newman, 2009; DVJ Insights and ING

The Netherlands, 2015) that journalists spend a con-

siderable amount of their time scouting social media

for interesting topics to write about, relying on these

platforms as more or less reliable sources.

It is therefore relevant to study how events or news

come about on these two types of platforms — news

articles and social posts. Of particular interest is the

issue of how and when they arise, disseminate, gain

strength and die. The main objective of our work is to

focus on the timing of their generation and the inten-

sity with which they occur in both media and exploit

text mining techniques for this aim.

In this preliminary study, we are especially inter-

ested in comparing press publication moments with

the community discussion moments, in social media,

for different kinds of topics: longstanding, entity-

oriented, and event-oriented. What are the new tem-

poral trends? Are events ﬁrst discussed in social me-

dia, becoming later main-stream media, or still in the

other way around? Understanding these temporal dy-

500

Moutinho, V., Brazdil, P. and Cordeiro, J.

Association and Temporality between News and Tweets.

DOI: 10.5220/0008362105000507

In Proceedings of the 11th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2019), pages 500-507

ISBN: 978-989-758-382-7

namics are a crucial issue for many other ﬁelds. In

our ﬁndings, we have observed that the trend in the

origin of news stories is shifting from traditional me-

dia to social media, at least for certain event-oriented

topics.

2 RELATED WORK

The literature concerning Twitter and news is often

directed at using tweets as a single news source. In-

deed, there are some studies where Twitter is regarded

as a substitute of (rather than complementary platform

to) traditional news sources (e.g. (Sankaranarayanan

et al., 2009; Zhao et al., 2011; Phuvipadawat and Mu-

rata, 2010)). The main reason for this is perhaps the

realization that some news break ﬁrst on Twitter. For

example, (Hu et al., 2012) have shown that the cap-

ture and death of Osama Bin Laden was made public

on Twitter at least 20 minutes sooner than on major

U.S. television channels. The authors argue that this

may happen due to the role of a particular set of inﬂu-

ential users, namely journalists and politicians, whose

credibility instantly leads to an immediate reaction on

social networks.

Sankaranarayanan et al. (Sankaranarayanan et al.,

2009) built a tool called TwitterStand with the goal of

collecting and diffusing breaking news quicker than

conventional news media. This system performs on-

line clustering on ﬁltered tweets from a set of manu-

ally selected seeders — users that usually post news.

In addition, it performs periodic checks to avoid frag-

mentation and ensure minimal duplication of clusters.

Also, it takes advantage of information in the content

of the tweet and/or the user’s proﬁle to associate top-

ics to geographic locations. The authors believe that

if tweets belonging to a certain cluster mostly come

from one location or a set of close locations, then the

topic of that cluster is likely to pertain to that geo-

graphical area.

Zhao et al. (Zhao et al., 2011) used a corpus

of news articles from the journal New York Times

(NYT) and tweets from users in Edinburgh, gath-

ered from November 11, 2009 to February 1, 2010,

to investigate how similar the topics in Twitter and

a traditional news source are. Their results showed

some differences regarding the most frequent cate-

gories and types of topics: Twitter users tweet mainly

about family and life, a category not covered by the

NYT; arts is a topic similarly frequent on both Twit-

ter and the NYT; world is much more frequent on

the NYT; lastly, while longstanding topics have an

equally strong presence, the same does not happen for

entity-oriented and event-oriented topics, with Twit-

ter favoring the former and the NYT the latter. Re-

garding long-lasting topics, there is evidence that their

prevalence is not due to an increasing number of users

tweeting about them, but to a set of important users

who discuss it over time (Kwak et al., 2010).

The above ﬁndings bring about relevant aspects

of the similarities between Twitter and conventional

news sources. Particularly they emphasize the impor-

tance of a certain type of users in social networks that

foster their role as a news medium. Still, while the

reputation and popularity of users is relatively high

regarding new information (Hu et al., 2012), the com-

munication structure set upon follower/followee rela-

tionships makes Twitter a fast information diffusion

network. In fact, this propagation may in some cases

not depend entirely on the ﬁrst user’s network: (Kwak

et al., 2010) have found that if a message is retweeted,

it quickly reaches an average of 1,000 users, regard-

less of the number of followers of the ﬁrst user. This is

what the authors call “the emergence of collective in-

telligence”, in the sense that individuals decide what

information is good enough to spread and once that

decision is made, it almost instantly reaches a mas-

sive audience.

None of the above works studied the issue of tem-

porality between news items and Twitter posts. This

motivated us to pursue this issue and report our ﬁnd-

ings in this article.

3 METHODOLOGY

The methodology used here is illustrated in Fig. 1.

It involves four main stages, from data gathering to

association and temporality analysis.

First, tweets and news are collected from news

sources and social media. Then, the news items are

clustered into groups of similar news. Afterwards,

tweets about the same news are associated with the

corresponding news clusters. Finally, the association

between news and tweets are examined, both regard-

ing the subject and temporality. The most relevant

parts of this process are detailed in the following sub-

sections.

3.1 Description of the Collected Data

The same story or event can be published in many

different articles and shared or commented in many

tweets. Therefore, a set of Portuguese news arti-

cles and tweets, from the ﬁrst semester of 2016, was

selected from the POPmine plat-form, developed in

SAPO Labs, a large database of news articles and so-

cial media content, automatically crawled from the

Association and Temporality between News and Tweets

501

Figure 1: The four main stages of our methodology.

Web (Saleiro et al., 2015).

To fulﬁll our research objectives, we targeted our

data collection on a sample of news stories containing

some degree of discussion on Twitter. This sample

comprises news stories directly referenced in Twitter,

through speciﬁc URL pointing to that story, as well as

those without any direct link. We are assuming that

if a tweet contains a link to a news article, then that

tweet is about the same story, naturally meaning that

in terms of temporal order the story came out ﬁrst on

mainstream news media. But in this study, we also

aim to measure the likelihood of the inverted tempo-

ral order, a new phenomenon with the emergence of

social media. That is, estimating the number of events

and stories that break out ﬁrst on social media, be-

coming mainstream media later.

Thus, the ﬁnal sample of data contains 6,074 news

articles and 11,328 tweets, from the same period (the

ﬁrst semester of 2016), eventually associated with

those news articles. These were selected from an orig-

inal set of near 600 thousand elements, observing the

following three criteria:

• We have observed a prevalence of sports topics in

the news articles from the considered period. In

order to obtain a wider range of topics we have

decided to focus only on generalist news sources.

• Despite using document vectorization with TF-

IDF and cosine similarity, we noticed that longer

articles are more likely to be grouped together, as

they contain more features, promoting thus doc-

ument similarity. After several clustering trials,

we have decided to only admit documents hav-

ing a length in the range of 100 to 3,349 charac-

ters. The upper bound of 3,349 enables the exclu-

sion of upper outlier documents in terms of doc-

ument length. The lower bound of 100 charac-

ters is used to exclude rather short and uninfor-

mative articles, such as the following examples:

“Dados s

ao relativos

a zona euro e

a Uni

ao Eu-

ropeia em geral.”;“Veja na

ıntegra o debate en-

tre os tr

es candidatos presidenciais, transmitido

na SIC Not

ıcias.”. It also prevents some articles

that may not have been fully or correctly collected

from entering the sample.

• The ﬁnal criterion for selecting the articles was

to include both articles that were shared on Twit-

ter and also articles that were not shared. Hence,

50% of the ﬁnal sample is comprised of online

news articles whose URL was shared in at least

one tweet and 50% of online news articles whose

URL was not found in any of the tweets. There

were 3,037 online news articles with a link to at

least one tweet. The other 50% was randomly as-

sembled, resulting in a ﬁnal sample of 6,074 arti-

cles.

Most of the original 19.4 million Portuguese

tweets from the ﬁrst semester of 2016 were discarded,

as they report more on personal, family and life sub-

jects, and so irrelevant for our study, as they are not

associated with any new stories. The selection of

tweets potentially relevant for our study was done as

follows. First, 5,664 tweets were selected based on

the presence of an URL to one of the previously se-

lected news articles. Secondly, an equal number of

tweets without any URL were also selected, follow-

ing a process that was both random and con-trolled.

First, 250 thousand tweets with more than 20 charac-

ters were randomly chosen. Then, only those contain-

ing at least one keyword from the clusters (see Section

3.2) of previously selected news, were kept (approxi-

mately 50%). Final-ly, a random subsample of these

was selected. The decision of this selection process

was therefore a compromise between the processing

resources available and the identiﬁcation of promis-

KDIR 2019 - 11th International Conference on Knowledge Discovery and Information Retrieval

502

ing tweets.

Standard pre-processing techniques were applied

to both news articles and tweets, including tokeniza-

tion, lower case conversion, punctuation and numbers

removal, Portuguese stopwords removal and stem-

ming. Additionally, we applied parts-of-speech tag-

ging to identify and keep verbs, nouns or proper

nouns. The recognition of named entities, such as

personalities, names of events and locations, was also

included, as we expected these features to help in

the representation and pattern recognition of groups

of stories or events. This task was done using the

PAMPO method (Rocha et al., 2016), built for the

Portuguese language.

3.2 News Clustering and Labeling

To obtain groups of similar stories clustering tech-

niques were applied to the selected sample of online

news articles. We have identiﬁed similar news arti-

cles and tweets and characterized each group with a

number of keywords.

In terms of clustering, the hybrid buckshot method

(Cutting et al., 1992) was used, which combines hi-

erarchical clustering with k-means. The initial cen-

ters for the k-means clustering are chosen by ﬁrst run-

ning the hierarchical clustering on a sample of

√

k ·N

news articles, where N represents the number of ar-

ticles, and k the number of centroids. The ideal k

value was chosen based on the representation of the

aggregation indices of the hierarchical clustering and

the representation of the explained inertia, which can

be expressed as a ratio B/T. The term B represents

between-class dispersion, and is measured as the sum

of squared distances of the cluster centers (centroids)

to the center of gravity g. The term T represents the

total dispersion of the data, and is measured as the

sum of squared distances of every observation to the

center of gravity g:

B =

∑

h=1

d(g

, g)

T =

∑

i=1

d(I

, g)

(1)

The explained inertia (B/T) naturally increases with

the number of clusters. The goal is to ﬁnd the value

of k for which this value starts to marginally decrease

(the elbow method) (Bholowalia and Kumar, 2014).

In our case this was around k=50. The clusters iden-

tiﬁed exhibit non-uniform distribution of documents,

some with large numbers and others with only a few.

The groups of news articles were labeled using

the most signiﬁcant terms as keywords. The num-

ber of keywords varies according to the cluster size,

so that larger clusters were represented by a larger

set of keywords. Keywords were ordered accord-

ing to their average TF-IDF values within the clus-

ter. As an example, Fig. 2 shows clusters of (a) ar-

ticles related to the Brexit referendum, (b) the ﬁrst

news about the run of Ant

onio Guterres’ for United

Nations Secretary-General, (c) the European Football

Cup held in France and (d) the Brussels terrorist at-

tack. The names of the clusters were given after an

examination of the list of keywords. We highlight

the importance of named entities identiﬁcation in this

study, performed using PAMPO (Rocha et al., 2016),

since terms such as Reino Unido (United Kingdom),

Uni

ao Europeia (European Union), Nac¸

oes Unidas

(United Nations) and Secret

ario-Geral (Secretary-

General) are often very indicative of the cluster’s

main subject.

In this work we have selected a subset of four clus-

ters for further analysis, based on their size, the topic

addressed and the number of associated tweets. They

were the Brexit, European Football Cup and Air trans-

ports incidents presented above, and also a cluster on

Politics.

3.3 Assigning Tweets to Clusters

In order to analyse when a certain text or group of

similar texts appear in social networks in comparison

to its press release, there is a need to assign social me-

dia posts to a particular group of similar texts. In this

case, we used a collection of tweets posted during the

same period as the online news articles and assigned

them to the appropriate clusters of articles.

Tweets were assigned to the clusters of news arti-

cles using the cosine similarity measure between each

tweet and the cluster centroids, computed using fea-

ture vectors with normalized TF-IDF values. For each

tweet, the ﬁve closest cluster centroids were identi-

ﬁed, and the cluster that was most similar to the tweet

in question was chosen. The sample of tweets is larger

than the sample of articles, and the mean ratio is 1.8

tweets per article. For the clusters under observa-

tion, the ratio is larger than the mean ratio (from 2.1

for cluster Brexit, to 3.9 for cluster Air Transport in-

cidents), with the exception of cluster Politics (0.8).

This is some evidence that the chosen clusters have a

place in the social network discussion.

3.3.1 Assignment

As all the data used in this study are unlabeled, re-

sults cannot be directly evaluated. We addressed this

issue by including tweets with links to clustered news

articles. It can be assumed that if a tweet shares a spe-

ciﬁc news article, it should belong to the same clus-

ter. So, we consider the cluster as the real class of

Association and Temporality between News and Tweets

503

Figure 2: Keyword clouds for four news clusters.

the tweet and compare it with the results of the as-

signment based on similarity to cluster centroids. Ac-

curacy, precision, recall and F1 measures were used

to evaluate these results. In addition, we borrowed

the concept of precision at n (P@n) from the infor-

mation retrieval ﬁeld (Sch

utze et al., 2008). In this

study, we made the following adaptation: each obser-

vation leads to n predictions, based on the distance to

the closest news centroids. If the true class of a tweet

is in the n-topmost predictions, it is considered as a

true positive. The P@n is therefore the percentage of

observations whose true class is present in the top n

predictions.

We present the values for P@n, for n up to ﬁve

(see Table 1). In this case, P@5 is 43.3%, which

means that, for 43.3% of the assigned tweets with link

to a news article, the correct cluster was in the top ﬁve

predictions.

Regarding the detailed analysis of our subset of clus-

ters, Brexit has the highest precision value (43.2%)

(see Table 2). However, the value of F1 is higher for

Air transport incidents (39.7%), Football - Euro 2016

(37.0%), and Brexit is in third place (34.7%). These

performance values point to the clusters that are prob-

ably the most reliable for the temporality analysis dis-

cussed in the next section.

Table 1: Global evaluation of tweets assignment to clusters.

Measure Value P@n Value

Accuracy 12.7% P@1 14.2%

MAV Precision 13.8% P@2 21.5%

MAV Recall 33.4% P@3 29.2%

MAV F1 14.7% P@4 36.1%

P@5 43.3%

Table 2: Performance of the selected subset of clusters.

Cluster Precision Recall F1

Air trans. inc. 27.4% 72.2% 39.7%

Foot. Euro 2016 27.4% 57.0% 37.0%

Brexit 43.2% 28.9% 34.7%

Politics 29.7% 1.2% 2.3%

4 TEMPORAL ANALYSIS

The main goal of this work was to identify similar

stories or events and analyze when they come about

on the news and social media. Given the rise of user

generated content and the current trend of journalists

scouting social media for crowd checking and news

monitoring (DVJ Insights and ING The Netherlands,

2015), the hypothesis is that these two environments

are interconnected. In this section we present the

timeline for all documents in each cluster, in order

to gain insights into the evolution of news generation

and sharing/commenting on Twitter in Portugal.

For this analysis we focused both on tweets that

do not have any link to a news article, as well as on

news articles that were not shared on Twitter. This

prevents a possible bias towards the hypothesis that,

for a given cluster, the discussion on Twitter occurred

after its publication by the press.

Fig. 3 presents the temporal evolution of tweets

and articles for the four clusters under observation.

We recall that Football - Euro 2016, Brexit and Air

transport incidents (includes Brussels attack) were

the clusters with the best F1 scores on the evaluation

of tweets assignment, and that Politics is the largest

cluster, albeit with a lower performance evaluation

(see Table 2).

It is possible to observe that clusters Football - Euro

2016 and Brexit have peaks in the number of articles

KDIR 2019 - 11th International Conference on Knowledge Discovery and Information Retrieval

504

Figure 3: Evolution of the number of elements.

and tweets at the expected moment: the European

Football Cup started on the 10th of June (week 24)

and the Brexit Referendum took place on the 23rd of

June (week 26). Naturally, both of these events were

subject to discussion in the previous months, as the

National team prepared for the competition and de-

bates concerning Brexit and its consequences inten-

siﬁed. Tweets assigned to Football - Euro 2016 al-

ways surpassed the number of published articles on

a weekly basis, with a global ratio of six tweets to

one news article. This highlights the importance of

this event (Euro 2016) and topic (football) on the Por-

tuguese discussions on Twitter. These time series

show a correlation of 0.84, which may indicate that

football is referred to with the same intensity in the

news and on Twitter. A smoother trend of tweets sur-

passing the number of articles is noticed for Brexit,

with the exception of week 26, when the referendum

occurred, where the number of assigned tweets is ap-

proximately 50% lower than the clustered news arti-

cles.

Air transport incidents (includes Brussels attack)

is the smallest cluster under observation, albeit having

scored the highest F1 value. It shows a small rise at

week 13, which was when the bombings in Brussels

Airport and Maalbeck metro station happened. We re-

mark, nevertheless, that if we included shared articles

and tweets with link to them in this analysis, the rise

at week 13 would be signiﬁcantly larger (18 tweets

and 10 articles versus an average of 0.2 and 1.6 in the

weeks prior to this event).

Politics, the largest cluster of articles and news,

shows a rather smooth evolution in the number of el-

ements, especially on the tweets side. It shows that

for the kind of study further partitioning of this clus-

ter would be desirable in order to identify patterns in

texts possibly different from the ones revealed. This is

the cluster with the lowest ratio of assigned tweets to

articles, which may also be a sign that for this topic,

the keywords generated at the articles level may not

be sufﬁciently discriminative at the tweets level. The

evaluation measures did, indeed, reveal a large pro-

portion of false negatives for this class. Another pos-

sible line of interpretation is that Portuguese Twitter

users do not, in fact, talk as much about politics when

compared to its importance to the press.

4.1 Time-wise Differences

One way of analyzing the temporality between news

and tweets is to consider the time difference of ev-

ery article in a cluster to its median tweet. The me-

dian tweet is the tweet with median time in the cor-

responding cluster. This brings out a question of how

the press publication timings compare to the moment

when the public discussion is at its highest. The time

difference is computed as a lag variable, that, if posi-

tive, indicates that the article is older than the median

tweet, and the opposite if negative. If the distribution

of this variable is skewed to the right, the stories of

that cluster have a tendency to be ﬁrst published by

the press; if skewed to the left, the social media dis-

cussion happens sooner than the news. Fig. 4 shows

the representation of the distribution of this lag vari-

Association and Temporality between News and Tweets

505

Figure 4: Days difference between articles and the median tweet.

able for the considered subset of clusters.

It can be observed that the majority of the articles be-

longing to the cluster Football - Euro 2016 were pub-

lished after the median date of the tweets assigned to

this cluster (a negative days difference). Portuguese

Twitter users seem, therefore, to anticipate the discus-

sion of the national team participation in the compe-

tition in comparison to what happened in the news. A

similar conclusion can be drawn for Brexit: the height

of discussion of Portuguese users of the staying or

leaving of the UK from the European Union happened

on Twitter about two months before it happened in the

press.

These observations lead to the following conclu-

sion: while the news on Football - Euro 2016 and

Brexit were more event-oriented, with peaks of arti-

cles at speciﬁc points or short periods of time (e.g.:

the start of the football competition on the 10th of

June; the referendum date announcement in Febru-

ary and the referendum itself on the 23rd of June),

exchange of tweets has happened more evenly dis-

tributed during the period under study.

The Politics cluster does not present any speciﬁc

pattern. The Air transport incidents cluster shows

some signs that the press published articles about the

incidents ﬁrst — in total there were nine articles pub-

lished before the median tweet (positive days dif-

ference) and ﬁve articles published after the median

tweet (negative days difference).

We also did this temporality analysis for the whole

dataset, i.e., including shared articles and tweets with

links to them. Conclusions for the selected clusters

have not changed.

5 CONCLUSIONS

This work aimed at providing some insights into the

temporal relationship of texts that appear both as news

articles and in social networks in Portugal. The strat-

egy was to investigate what were the main themes

published and how they behaved in terms of the num-

ber of articles and social media posts during a reason-

ably long period.

To that end, a sample of online news articles

from generalist news sources was subject to text clus-

tering techniques, and 50 groups of similar articles

were identiﬁed. These were subsequently labelled

with keywords. The groups included articles on foot-

ball related events and teams, terrorist attacks and

other international incidents, elections, accidents, in-

vestigations, political parties, ministerial actions, eco-

nomic/ﬁnancial reports and weather warnings, among

others.

Then, to associate tweets with the groups of news

articles, we used a sample of tweets and assigned

them to the clusters using a similarity measure. This

assignment was evaluated using tweets with a news

article URL. The fact that tweets contain very few

terms makes this task rather difﬁcult. Also, as the

number of classes is large, we cannot expect a very

high accuracy. The default accuracy for 50 classes

KDIR 2019 - 11th International Conference on Knowledge Discovery and Information Retrieval

506

with equal distribution of examples in training would

be 1/50, i.e. only 2%. The accuracy achieved was

12.7%, that is substantially higher than the default.

Four clusters for which the evaluation was con-

siderably above average, were chosen for the sub-

sequent study of the temporality between news and

tweets. These were Brexit, Football - Euro 2016, Air

transport incidents and Politics. Regarding Football

- Euro 2016 the football national team was a rather

constant subject of discussion on social media in the

ﬁrst semester of 2016, culminating in the ﬁnal month

with its participation in the European Cup. However,

the same did not happen on the news side, where

the most frequent articles were published towards the

end of the period under observation. A similar pat-

tern was observed for Brexit. This indicates that, for

some texts, the press is more event-oriented, contrast-

ing with the more permanent focus of Twitter users.

The analysis Air transport incidents, which included

Brussels bombings, revealed that the press had a more

prominent role in the news diffusion, while comments

in social media appeared afterwards. Regarding the

cluster Politics, no pattern was identiﬁed. Perhaps,

with a more reﬁned level of partitioning certain tim-

ing patterns would be more easily identiﬁed.

Regarding future work, we believe that the results

could be improved by adopting ontologies that would

enable to compute semantic distances between arti-

cles and tweets. Besides, tweets could also be ex-

panded with the use of synonyms or through word

embeddings (Mikolov et al., 2013). This would en-

hance the matching process of tweets to the lexicon

obtained from the articles.

ACKNOWLEDGMENT

This work has been supported by the project Centro-

01-0145-FEDER-000019 - C4 – Cloud Computing

Competences Centre” coﬁnanced, through the Sup-

port System for Scientiﬁc and Technological Re-

search - Integrated SR&TD Programs, by the Portugal

2020 Program (PT 2020), in the framework of the Re-

gional Operational Program of the Center (CENTRO

2020) and by the European Union through the Euro-

pean Regional Development Fund (ERDF).

REFERENCES

Bholowalia, P. and Kumar, A. (2014). EBK-means: A clus-

tering technique based on elbow method and k-means

in WSN. International Journal of Computer Applica-

tions, 105(9):17–24.

Cutting, D. R., Karger, D. R., Pedersen, J. O., and Tukey,

J. W. (1992). Scatter/Gather: A Cluster-based Ap-

proach to Browsing Large Document Collections. In

SIGIR 92, volume 51, pages 318–329, Copenhagen,

Denmark. ACM.

DVJ Insights and ING The Netherlands (2015). Impact of

social media on news (#SMING15). Technical report.

Hu, M., Liu, S., Wei, F., Wu, Y., Stasko, J., and Ma, K.-

L. (2012). Breaking news on twitter. In Proceedings

of the 2012 ACM annual conference on Human Fac-

tors in Computing Systems - CHI ’12, pages 275–279,

Austin, Texas, USA. ACM.

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.

Kwak, H., Lee, C., Park, H., and Moon, S. (2010). What is

Twitter, a Social Network or a News Media? In WWW

’10 Proceedings of the 19th International Conference

on World Wide Web, pages 591–600, Raleigh, North

Carolina, USA. ACM.

Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., and

Dean, J. (2013). Distributed representations of words

and phrases and their compositionality. In Advances in

neural information processing systems, pages 3111–

3119.

Newman, N. (2009). The rise of social media and its impact

on mainstream journalism. Technical Report Septem-

ber, Reuters Institute for the Study of Journalism, De-

partment of Politics and International Relations, Uni-

versity of Oxford.

Phuvipadawat, S. and Murata, T. (2010). Breaking news de-

tection and tracking in Twitter. In Proceedings - 2010

IEEE/WIC/ACM International Conference on Web In-

telligence and Intelligent Agent Technology - Work-

shops, WI-IAT 2010, pages 120–123, Toronto, ON,

Canada. IEEE.

Rocha, C., Jorge, A., Sionara, R., Brito, P., Pimenta, C., and

Rezende, S. (2016). PAMPO: using pattern matching

and pos-tagging for effective Named Entities recogni-

tion in Portuguese. arXiv preprint arXiv:1612.09535.,

pages 1–17.

Saleiro, P., Amir, S., Silva, M., and Soares, C. (2015).

POPmine: Tracking Political Opinion on the Web.

In 2015 IEEE International Conference on Computer

and Information Technology; Ubiquitous Computing

and Communications; Dependable, Autonomic and

Secure Computing; Pervasive Intelligence and Com-

puting (CIT/IUCC/DASC/PICOM), pages 1521–1526,

Liverpool, UK. IEEE.

Sankaranarayanan, J., Samet, H., Teitler, B. E., Lieberman,

M. D., and Sperling, J. (2009). TwitterStand: News in

Tweets. In Proceedings of the 17th ACM SIGSPATIAL

International Conference on Advances in Geographic

Information Systems - GIS ’09, pages 42–51, Seattle,

Washington, USA. ACM.

Sch

utze, H., Manning, C. D., and Raghavan, P. (2008).

Evaluation in information retrieval, volume 39. Cam-

bridge University Press, New York, USA, 1st edition.

Zhao, W. X., Jiang, J., Weng, J., He, J., Lim, E.-p., Yan,

H., and Li, X. (2011). Comparing Twitter and Tradi-

tional Media using Topic Models. In Proceedings of

the 33rd European conference on Advances in infor-

mation retrieval (ECIR’11), pages 338–349. Springer,

Berlin, Heidelberg.

Association and Temporality between News and Tweets

507