Complex & Intelligent Systems,
Journal Year:
2024,
Volume and Issue:
10(3), P. 3763 - 3788
Published: Feb. 23, 2024
Abstract
Network
embedding
is
a
technique
used
to
generate
low-dimensional
vectors
representing
each
node
in
network
while
maintaining
the
original
topology
and
properties
of
network.
This
technology
enables
wide
range
learning
tasks,
including
classification
link
prediction.
However,
current
landscape
approaches
predominantly
revolves
around
static
networks,
neglecting
dynamic
nature
that
characterizes
real
social
networks.
Dynamics
at
both
micro-
macrolevels
are
fundamental
drivers
evolution.
Microlevel
dynamics
provide
detailed
account
formation
process,
macrolevel
reveal
evolutionary
trends
Despite
recent
efforts,
few
accurately
capture
evolution
patterns
nodes
microlevel
or
effectively
preserve
crucial
layers.
Our
study
introduces
novel
method
for
i.e.,
bilayer
pattern-preserving
networks
(Bi-DNE),
preserves
macrolevels.
The
model
utilizes
strengthened
triadic
closure
represent
structure
process
microlevel,
equation
constrains
adhere
densification
power-law
pattern
macrolevel.
proposed
Bi-DNE
exhibits
significant
performance
improvements
across
prediction,
reconstruction,
temporal
analysis.
These
demonstrated
through
comprehensive
experiments
carried
out
on
simulated
real-world
datasets.
consistently
superior
results
those
state-of-the-art
methods
empirical
evidence
effectiveness
capturing
complex
high-quality
representations.
findings
validate
methodological
innovations
presented
this
work
mark
valuable
progress
emerging
field
representation
learning.
Further
exploration
demonstrates
sensitive
analysis
task
parameters,
leading
more
accurate
natural
during
embedding.
IEEE Transactions on Knowledge and Data Engineering,
Journal Year:
2024,
Volume and Issue:
36(11), P. 6989 - 7002
Published: April 16, 2024
Attributed
graph
clustering
aims
to
group
nodes
into
disjoint
categories
using
deep
learning
represent
node
embeddings
and
has
shown
promising
performance
across
various
applications.
However,
two
main
challenges
hinder
further
improvement.
Firstly,
reliance
on
unsupervised
methods
impedes
the
of
low-dimensional,
clustering-specific
features
in
representation
layer,
thus
impacting
performance.
Secondly,
predominant
use
separate
approaches
leads
suboptimal
learned
that
are
insufficient
for
subsequent
steps.
To
address
these
limitations,
we
propose
a
novel
method
called
Semi-supervised
Deep
Clustering
Dual
Autoencoder
(SDAC-DA).
This
approach
enables
semi-supervised
end-to-end
attributed
networks,
promoting
high
structural
cohesiveness
attribute
homogeneity.
SDAC-DA
transforms
network
dual-view
network,
applies
autoencoder
layering
each
view,
integrates
dimensionality
reduction
matrices
by
considering
complementary
views.
The
resulting
layer
contains
clustering-friendly
embeddings,
which
optimized
through
unified
process
effectively
identifying
clusters.
Extensive
experiments
both
synthetic
real
networks
demonstrate
superiority
our
proposed
over
seven
state-of-the-art
approaches.
Journal of King Saud University - Computer and Information Sciences,
Journal Year:
2023,
Volume and Issue:
35(10), P. 101798 - 101798
Published: Oct. 16, 2023
The
influence
of
the
node
refers
to
ability
disseminate
information.
faster
and
wider
spreads,
greater
its
influence.
There
are
many
classical
topological
metrics
that
can
be
used
evaluate
influencing
nodes.
Degree
centrality,
betweenness
closeness
centrality
local
among
most
common
for
identifying
influential
nodes
in
complex
networks.
is
very
simple
but
not
effective.
Global
such
as
better
identify
nodes,
they
compatible
on
large-scale
networks
due
their
high
complexity.
In
order
design
a
ranking
method
this
paper
new
semi-local
metric
proposed
based
relative
change
average
shortest
path
entire
network.
Meanwhile,
our
provides
quantitative
global
importance
model
measure
overall
each
node.
To
performance
metric,
we
use
Susceptible-Infected-Recovered
(SIR)
epidemic
model.
Experimental
results
several
real-world
show
has
competitive
with
existing
equivalent
efficiency
dealing
effectiveness
been
proven
numerical
examples
Kendall's
coefficient.
IEEE Transactions on Computational Social Systems,
Journal Year:
2023,
Volume and Issue:
11(3), P. 3444 - 3456
Published: Nov. 8, 2023
Network
clustering
is
one
of
the
fundamental
unsupervised
methods
knowledge
discovery.
Its
goal
to
group
similar
nodes
together
without
supervision
or
prior
nature
clusters.
Among
various
methods,
semi-supervised
detection
most
promising
approaches
for
community
because
its
ability
employ
side
information
better
understand
network
topology.
However,
previous
work
faces
two
problems:
use
linear
reduce
dimensionality
and
random
selection
information,
as
a
result
these
drawbacks,
are
less
efficient.
To
fill
gaps,
we
developed
an
end-to-end
deep
semi-supervisor
(DSSC)
complex
networks.
A
new
learning
objective
designed
that
uses
semi-autoencoder
(SeAE)
with
defined
pair-wise
constraint
matrix
based
on
point-wise
mutual
(PMI)
in
representation
layer
accurately
learn
distinctive
features
and,
layer,
adds
term
minimize
distance
within
cluster
while
between
clusters
increases.
The
results
show
our
method
performs
unexpectedly
well
comparison
existing
state-of-the-art
Neural Computing and Applications,
Journal Year:
2023,
Volume and Issue:
35(34), P. 24493 - 24511
Published: Oct. 3, 2023
Abstract
Attributed
graph
clustering,
the
task
of
grouping
nodes
into
communities
using
both
structure
and
node
attributes,
is
a
fundamental
problem
in
analysis.
Recent
approaches
have
utilized
deep
learning
for
embedding
followed
by
conventional
clustering
methods.
However,
these
methods
often
suffer
from
limitations
relying
on
original
network
structure,
which
may
be
inadequate
due
to
sparsity
noise,
separate
that
yield
suboptimal
embeddings
clustering.
To
address
limitations,
we
propose
novel
method
called
Deep
Clustering
with
High-order
Proximity
Preserve
(DAC-HPP)
attributed
DAC-HPP
leverages
an
end-to-end
framework
integrates
high-order
proximities
fosters
structural
cohesiveness
attribute
homogeneity.
We
introduce
modified
Random
Walk
Restart
captures
k-order
information,
enabling
modelling
interactions
between
proximities.
A
consensus
matrix
representation
constructed
combining
diverse
proximity
measures,
joint
approach
employed
leverage
complementary
strengths
In
summary,
offers
unique
solution
incorporating
employing
framework.
Extensive
experiments
demonstrate
its
effectiveness,
showcasing
superiority
over
existing
Evaluation
synthetic
real
networks
demonstrates
outperforms
seven
state-of-the-art
approaches,
confirming
potential
advancing
research.
Multimedia Tools and Applications,
Journal Year:
2024,
Volume and Issue:
83(24), P. 64795 - 64811
Published: Jan. 17, 2024
Abstract
The
facilitation
of
sharing
and
exchanging
patients’
health
records
is
a
paramount
opportunity
in
e-health,
enabling
healthcare
providers
to
garner
comprehensive
clear
perspective
medical
histories
without
necessitating
direct
inquiries.
Besides
this
great
advantage,
it
introduces
substantial
issues
on
security
privacy,
mainly
related
unauthorized
access
e-health
when
different
service
maintain
records.
In
paper,
we
deal
with
problem
propose
using
the
blockchain
technology
(1)
obfuscate
linkage
between
identities
their
(2)
grant
exclusively
entities
authorized
by
patients
themselves.
Key
outcomes
include
digital
identity
based
Electronic
Identification,
Authentication,
Trust
Services
Regulation
(eIDAS)
control
these
records,
concrete
implementation
adopting
Ethereum
blockchain.
Our
solution
relies
public
blockchain,
which
an
improvement
for
state
art,
only
private
or
consortium
blockchains
have
been
proposed.
resulting
has
analyzed,
effectiveness
affordability
proposal
shown.
