IEEE Transactions on Affective Computing,
Journal Year:
2023,
Volume and Issue:
14(4), P. 2564 - 2566
Published: Oct. 1, 2023
In
The
formative
years
of
Affective
Computing
[1],
from
the
late
1990s
and
into
early
2000s,
a
significant
fraction
research
attention
was
focused
on
development
methods
for
unobtrusive
physiological
measurement
.
It
quickly
became
obvious
that
wiring
people
with
electrodes
strapping
cumbersome
hardware
to
their
bodies
not
only
restricting
types
experiments
could
be
performed
but
also
conducive
unbiased
observations.
For
instance,
subjects
fingers
wrapped
electrodermal
activity
(EDA)
photoplethysmography
(PPG)
sensors
hardly
type,
drive
or
sleep
comfortably.
Hence,
there
need
more
elegant
scalable
measurement
[2].
IEEE Transactions on Knowledge and Data Engineering,
Journal Year:
2024,
Volume and Issue:
36(7), P. 2956 - 2966
Published: Jan. 5, 2024
Depression
is
one
of
the
most
common
mental
illnesses,
but
few
currently
proposed
in-depth
models
based
on
social
media
data
take
into
account
both
temporal
and
spatial
information
in
for
detection
depression.
In
this
paper,
we
present
an
efficient,
low-covariance
multimodal
integrated
spatio-temporal
converter
framework
called
DepMSTAT,
which
aims
to
detect
depression
using
acoustic
visual
features
data.
The
consists
four
modules:
a
preprocessing
module,
token
generation
Spatial-Temporal
Attentional
Transformer
(STAT)
classifier
module.
To
efficiently
capture
correlations
data,
plug-and-play
STAT
module
proposed.
capable
extracting
unimodal
fusing
information,
playing
key
role
analysis
Through
extensive
experiments
database
(D-Vlog),
method
paper
shows
high
accuracy
(71.53%)
detection,
achieving
performance
that
exceeds
models.
This
work
provides
scaffold
studies
assists
IEEE Access,
Journal Year:
2024,
Volume and Issue:
12, P. 103976 - 104019
Published: Jan. 1, 2024
Emotion
recognition
involves
accurately
interpreting
human
emotions
from
various
sources
and
modalities,
including
questionnaires,
verbal,
physiological
signals.
With
its
broad
applications
in
affective
computing,
computational
creativity,
human-robot
interactions,
market
research,
the
field
has
seen
a
surge
interest
recent
years.
This
paper
presents
systematic
review
of
multimodal
emotion
(MER)
techniques
developed
2014
to
2024,
encompassing
signals,
facial,
body
gesture,
speech
as
well
emerging
methods
like
sketches
recognition.
The
explores
models,
distinguishing
between
emotions,
feelings,
sentiments,
moods,
along
with
emotional
expression,
categorized
both
artistic
non-verbal
ways.
It
also
discusses
background
automated
systems
introduces
seven
criteria
for
evaluating
modalities
alongside
current
state
analysis
MER,
drawn
human-centric
perspective
this
field.
By
selecting
PRISMA
guidelines
carefully
analyzing
45
selected
articles,
provides
comprehensive
perspectives
into
existing
studies,
datasets,
technical
approaches,
identified
gaps,
future
directions
MER.
highlights
challenges
Frontiers in Neurology,
Journal Year:
2024,
Volume and Issue:
15
Published: July 4, 2024
Introduction
Depressive
and
manic
states
contribute
significantly
to
the
global
social
burden,
but
objective
detection
tools
are
still
lacking.
This
study
investigates
feasibility
of
utilizing
voice
as
a
biomarker
detect
these
mood
states.
Methods:From
real-world
emotional
journal
recordings,
22
features
were
retrieved
in
this
study,
21
which
showed
significant
differences
among
Additionally,
we
applied
leave-one-subject-out
strategy
train
validate
four
classification
models:
Chinese-speech-pretrain-GRU,
Gate
Recurrent
Unit
(GRU),
Bi-directional
Long
Short-Term
Memory
(BiLSTM),
Linear
Discriminant
Analysis
(LDA).
Results
Our
results
indicated
that
Chinese-speech-pretrain-GRU
model
performed
best,
achieving
sensitivities
77.5%
54.8%
specificities
86.1%
90.3%
for
detecting
depressive
states,
respectively,
with
an
overall
accuracy
80.2%.
Discussion
These
findings
show
machine
learning
can
reliably
differentiate
between
via
analysis,
allowing
more
precise
approach
disorder
assessment.
Frontiers in Psychiatry,
Journal Year:
2025,
Volume and Issue:
15
Published: Jan. 23, 2025
Introduction
Depression
is
a
prevalent
mental
disorder,
and
early
screening
treatment
are
crucial
for
detecting
depression.
However,
there
still
some
limitations
in
the
currently
proposed
deep
models
based
on
audio-video
data,
example,
it
difficult
to
effectively
extract
select
useful
multimodal
information
features
from
very
few
studies
have
been
able
focus
three
dimensions
of
information:
time,
channel,
space
at
same
time
depression
detection.
In
addition,
challenges
utilizing
other
tasks
enhance
prediction
accuracy.
The
resolution
these
issues
constructing
Methods
this
paper,
we
propose
multi-task
representation
learning
vision
audio
detection
model
(DepITCM).The
comprises
main
modules:
data
preprocessing
module,
Inception-Temporal-Channel
Principal
Component
Analysis
Module(ITCM
Encoder),
module.
To
efficiently
rich
feature
representations
video
ITCM
Encoder
employs
staged
extraction
strategy,
transitioning
global
local
features.
This
approach
enables
capture
while
emphasizing
fusion
temporal,
spatial
finer
detail.
Furthermore,
inspired
by
strategies,
paper
enhances
primary
task
classification
incorporating
secondary
(regression
task)
improve
overall
performance.
Results
We
conducted
experiments
AVEC2017
AVEC2019
datasets.
results
show
that,
task,
our
method
achieved
an
F1
score
0.823
accuracy
dataset,
0.816
0.810
dataset.
regression
RMSE
was
6.10
(AVEC2017)
4.89
(AVEC2019),
respectively.
These
demonstrate
that
outperforms
most
existing
methods
both
tasks.
can
performance
when
using
learning.
Discussion
Although
through
multimodality
has
shown
good
previous
studies.
utilize
complementary
between
different
Therefore,
work
combines
Previous
mostly
focused
ignoring
importance
Based
problems
studies,
made
corresponding
improvements
provide
more
comprehensive
effective
solution
Frontiers in Human Neuroscience,
Journal Year:
2025,
Volume and Issue:
19
Published: Feb. 26, 2025
With
the
rapid
development
of
deep
learning,
Electroencephalograph(EEG)
emotion
recognition
has
played
a
significant
role
in
affective
brain-computer
interfaces.
Many
advanced
models
have
achieved
excellent
results.
However,
current
research
is
mostly
conducted
laboratory
settings
for
induction,
which
lacks
sufficient
ecological
validity
and
differs
significantly
from
real-world
scenarios.
Moreover,
are
typically
trained
tested
on
datasets
collected
environments,
with
little
validation
their
effectiveness
situations.
VR,
providing
highly
immersive
realistic
experience,
an
ideal
tool
emotional
research.
In
this
paper,
we
collect
EEG
data
participants
while
they
watched
VR
videos.
We
propose
purely
Transformer-based
method,
EmoSTT.
use
two
separate
Transformer
modules
to
comprehensively
model
temporal
spatial
information
signals.
validate
EmoSTT
passive
paradigm
environment
active
dataset
environment.
Compared
state-of-the-art
methods,
our
method
achieves
robust
classification
performance
can
be
well
transferred
between
different
elicitation
paradigms.
