IEEE Systems, Man and Cybernetics Magazine - April 2020 - 40

Islam and Chowdhury [139] presented a spam classification technique using a clustering approach to categorize
the features. They used the VAT algorithm to assess the
extracted features and then passed the information into a
classification engine (consisting of tree-based classifiers,
nearest-neighbor algorithms, statistical algorithms, and so
on). Cornelissen et al. [140] proposed that the social network topology of a user would be sufficient to determine
whether the user is an automated agent or a human. They
tested their conjecture on a publicly available data set containing users on Twitter labeled as either automated social
agents or humans. The VAT algorithm was used to determine the best distance measure, which provides the best
performance in classifying users.
Graph Data Analysis
A graph represents data consisting of nodes (representing objects with certain attributes) and the relationship
between different nodes (represented by edges between
pairs of nodes). Common examples of graph data include
social (people-people relationship), coauthorship, road,
power, and water networks, among others. Analyzing
graphs is useful for determining general trends, relating
the results of an experiment to the hypothesis, and formulating hypotheses for future investigations. Four
widely used types of graph analytics include path, connectivity, community, and centrality analyses. In this section, we describe some papers that have applied a member
of the VAT family of algorithms to various graph-data
analysis problems.
Community Detection
Community detection is one of the most popular topics of
modern network science. Communities, or clusters, are usually thought of as groups of vertices that have a higher probability of being connected to each other than to members of
other groups. However, identifying a community is an illdefined problem. There are no universal protocols for the
fundamental ingredients, such as the definition of the community itself, or for other crucial issues, including the validation of algorithms and comparison of their performances.
Yang et al. [141] were the first to use the VAT algorithm
to detect communities in a graph. The application of VAT
to graph data is not straightforward since there is not a
general meaningful distance in a graph. The authors introduced a new distance between nodes to measure the dissimilarity between nodes and obtain the distance matrix,
which was then reordered using VAT to detect the community structure hidden in complex networks.
An important challenge in many graph clustering applications is that the clusters are not crisp (graph nodes may
be partially associated with several clusters), leading to
fuzzy clusters in graphs. Runkler and Bezdek [142] proposed the use of relational fuzzy clustering-more specifically, NERF c-means-to the relational data (obtained
from the adjacency matrix of a graph). The clustering
40

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE Apri l 2020

results were visualized using VAT and were illustrated on
Zachary's karate-club benchmark data [27].
Papers by Havens et al. [143] and Su and Havens [144],
[145] used various approaches, such as genetic algorithms
and fuzzy modularity maximization for fuzzy community
detection in social network graph data. The Newman-Girvan (NG) modularity function that measures how vertices
in a community share more edges than expected in a randomized network were used as a cluster validity function.
All of these papers used VAT, iVAT, and SpecVAT to visualize the results of various clustering approaches.
Ganji et al. [146] proposed a generalized modularity
measure called GM, which has a more sophisticated interpretation of vertex similarity than vertex similarity as
measured by the NG modularity function. GM takes into
account the number of longer paths between vertices,
compared to what would be expected in a randomized network, something that the NG modularity function does not
consider. The VAT algorithm was used to illustrate how
well-generalized modularity can reveal the underlying
community structures in real-world graph data.
Visualizing Networks
Visualization of small-world networks is challenging owing
to the large size of the data and their property of being
"locally dense but globally sparse." Generally, networks
are represented using graph layouts and images of adjacency matrices, which have shortcomings of occlusion and
spatial complexity in direct form. Parveen and SreevalsanNair [29] proposed a technique to enable effective and efficient visualization of small-world networks in the similarity
space, as opposed to the attribute space, using a similarity
matrix representation. They used VAT seriation to perform
multilevel clustering on the matrix form and visualize a
series of similarity matrices from the same data set using
parallel-sets-like representation.
IoT and Smart Cities
The IoT infrastructure for the creation of smart cities consists
of Internet-connected sensors, devices, and citizens. This IoT
infrastructure generates an enormous amount of data in the
form of city-scale physical measurements and public opinions, constituting big data. Smart cities aim to efficiently use
this wealth of data to manage and solve the problems faced
by modern cities for better decision making. Interpreting the
massive amount of smart-city-generated big data to create
actionable knowledge is a challenging task. Many researchers
have utilized various algorithms belonging to the VAT family
to analyze smart-city-generated data to gain actionable
insights from them as discussed in the next section.
Smart City Urban Mobility
In an urban environment, high-quality mobility is a necessity for the success of other sectors and the creation of
jobs, and it plays a key role in cultivating an attractive environment for residents and businesses. With the increase in

IEEE Systems, Man and Cybernetics Magazine - April 2020

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - April 2020