Computational Intelligence - August 2016 - 19

A. Social Centrality Study

The analysis of an individual source, without our
The first example we consider here concerns the
study of the importance (in terms of centrality) of
knowledge integration framework, would have led
persons (politicians, television people, presenters,
to a different, less precise, social network.
guests), during the observation period.To perform
this analysis, we consider all the persons referred
stored in the knowledge graph as the weight of the edge connectby the tweets and Facebook posts associated to all episodes of the
ing each person to the People node, by the resource extractors.
TV show and build the underlying social network following the
Hence, to conduct this analysis, we only need to aggregate the
definition of TV event-based named entity similarity graph (see Definiweights of the out-edges of each person node. Within a single
tion 11 in Section VI-B). According to this definition, there is an
source the aggregation is performed by merging all social objects (a
edge between two person nodes if there exists a path between these
tweetset or a Facebook post) related to a given episode. Then, each
two persons, traversing a TV Event node. Interestingly, these paths
edge weight is multiplied by the total number of occurrences of the
may involve cross-source nodes, i.e., the analysis of an individual
concept node People. Finally, the cut-off method based on energy is
source, without our knowledge integration framework, would have
employed to filter out the less important entries. To consider the
led to a different, less precise, social network. Since more than one
popularity in both Twitter and Facebook as a whole, we merged the
tweetset and Facebook post may exist during the week associated to
Facebook post nodes and Tweetset nodes associated to each episode.
each episode, for each episode, the similarity graph is such that all the
The resulting weight for each person node i is then computed as
tweetsets and Facebook posts are merged to obtain an aggregated
episode representation. Moreover, each of them is associated to the

set of the most mentioned persons during the considered period.
w (i) all = a $ w (i) t + (1 - a) $ w (i) f , (1)
On our Twitter data, the above described analysis produced a
social network that we analyzed by computing the betweenness
where w (i) t, w (i) f and w (i) all are, respectively, the node weights
centrality [34], [35] of each node (i.e., the number of shortest
of the edge connecting i to the Tweetset node, the node weights
paths from all vertices to all others that pass through that node),
of the edge connecting i to the Facebook post node, and the
obtaining the results in Table 2 (a). These results show that Matresulting weight associated to the edge connecting i to the aggreteo Renzi is very central for this TV program. He is the Italian
gated social object node. In this experiment, we considered all
prime minister and he was strongly involved in each reform cited
sources with the same weight, i.e., a = 0.5.
before. It is important to note that he is the most central concept
in terms of betweenness even if he participated only in few epiTable 2 Top betweenness centrality scores of nodes from
different social networks.
sodes during the observation period. The following concept is
Matteo Salvini, that is one of the most active leaders of the cen(a) Twitter social network
ter-right coalition especially known for his stance against illegal
Rank
Person
Centrality
immigration and his strong criticism on the rules coming from
1
Matteo Renzi
0.5244
the European Commission about economy and immigration.
2
Matteo
Salvini
0.1624
The betweenness centrality computed for the Facebook
3
Silvio
Berlusconi
0.0477
data on people belonging to the knowledge graph is reported
in Table 2 (b). The most important concept in terms of central4
Murizio Landini
0.0379
ity is always Matteo Renzi, but, interestingly, his centrality
5
Elsa Fornero
0.0139
appears less dominant w.r.t. other people. This analysis shows
(b) Facebook social network
that one of the best ranked persons is Beppe Grillo. This is
Rank
Person
Centrality
probably due to the fact that Grillo's supporters are particularly
1
Matteo
Renzi
0.1662
active in this social media platform. Thus, in this social network,
2
Massimo Giannini
0.1550
the position of Grillo is more central than in the Twitter case.
By combining the two information sources, we may notice
3
Silvio Berlusconi
0.1159
that all the relevant information for both sources is preserved,
4
Matteo Salvini
0.0714
as shown by the ranked betweenness scores in Table 2 (c). In
5
Beppe Grillo
0.0655
particular, Matteo Renzi, the Prime Minister is still the most
(c) Combined social network
central concept and, almost all the most important Italian poliRank
Person
Centrality
tics actors are in the first positions of the ranking.
B. Popularity Study

The second experiment consists in computing the "episode popularity" of each person. The popularity of a given node is related to
the percentage of citations of the associated persons' names in tweets
and Facebook post comments. Notice that this information is

Matteo Renzi

0.3089

Matteo Salvini

0.1208

Silvio Berlusconi

0.1008

Massimo Giannnini

0.0689

Maurizio Landini

0.0362

AUGUST 2016 | IEEE Computational intelligence magazine

Table of Contents for the Digital Edition of Computational Intelligence - August 2016