Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10, N
o. 5,
O
ct
obe
r 202
0, pp.
4928
~
4936
IS
S
N: 20
88
-
8708, DO
I: 10
.11
591/ijece
.v1
0i5
.
pp
4928
-
49
36
4928
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
A
pri
or c
ase
stu
dy of n
atu
ra
l
l
an
guage
p
rocessin
g
on
diffe
rent do
m
ain
Shruthi
J
.
1
,
S
uma
Swam
y
2
1
Depa
rtment of
Com
pute
r
Scie
n
ce
and
Engi
ne
ering,
BMS
Instit
u
t
e
of Te
chnol
og
y
and
Mana
g
emen
t
,
Ind
ia
2
Depa
rt
m
ent
of C
om
pute
r
Scie
n
ce
and
Engi
ne
ering,
Sir
M
.
Visve
svara
y
a Instit
ut
e
of
T
ec
hno
log
y
,
India
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
un
22
, 2
01
8
Re
vised
Ma
r
14
,
2020
Accepte
d
Ma
r
2
4
, 202
0
In
the
pre
sen
t
s
ta
t
e
of
d
igi
t
al
world,
computer
m
ac
hine
do
not
under
stan
d
the
hum
an’
s
ordina
r
y
l
angua
g
e.
T
his
is
the
gre
at
bar
rie
r
be
twee
n
h
um
ans
and
digi
tal
s
y
stems
.
Henc
e,
rese
ar
che
rs
found
an
adva
nce
d
t
ec
h
nolog
y
th
at
provide
s
informati
on
to
th
e
users
f
rom
the
d
igi
tal
m
ac
hine.
How
eve
r,
nat
ura
l
la
nguag
e
proc
essing
(i.
e.
NLP)
is
a
bra
nch
of
AI
tha
t
has
signifi
ca
nt
implic
ation
on
t
he
wa
y
s
th
at
co
m
pute
r
m
ac
hine
and
hum
ans
can
int
er
ac
t
.
NLP
has
bec
om
e
an
essential
t
e
chnol
og
y
in
brid
ging
the
comm
unic
a
ti
on
gap
bet
wee
n
hum
ans
and
d
igi
t
al
data
.
Thus,
th
is
stud
y
p
rovide
s
the
n
ec
essit
y
of
the
NLP
in
the
c
urre
nt
computin
g
world
al
ong
w
it
h
diff
ere
nt
app
roa
che
s
and
the
ir
applic
at
ion
s.
It
al
so,
highl
ig
hts t
he
ke
y
ch
al
l
enge
s in
the
dev
el
opm
ent
of
new
NLP m
odel
.
Ke
yw
or
d
s
:
Ar
ti
fici
al
i
ntell
igence
Hu
m
an
l
an
gu
a
ge
Natu
ral
l
an
gu
a
ge
p
r
ocessi
ng
Text
m
ining
Copyright
©
202
0
Instit
ute of
Ad
v
ance
d
Engi
ne
eri
ng
and
Sc
ie
n
ce
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Sh
r
ut
hi J
.,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce a
nd
E
ng
i
ne
erin
g,
BM
S I
nst
it
ute o
f
Tech
nolo
gy
and Ma
na
gem
e
nt
,
Be
ng
al
uru,
I
ndia
.
Em
a
il
: shr
uth
i
j.resea
rch@gm
ai
l.com
1.
INTROD
U
CTION
Natu
rall
y,
hu
m
an
la
ngua
ge
is
com
plex
an
d
to
un
der
sta
nd
this
la
ngua
ge
,
a
syst
em
m
us
t
al
so
kn
ow
about
c
orres
pond
i
ng
gram
m
a
ti
cal
ru
le
s,
m
eanin
g
a
nd
c
onte
xt,
al
on
g
with
sla
ng
an
d
a
cronym
s
util
ized
i
n
the
la
ngua
ge.
Natu
ral
la
ngua
ge
processi
ng
(
NLP)
is
the
m
e
ch
anism
wh
ic
h
suppo
rts
the
c
om
pu
te
r
m
achine
by
si
m
ulati
ng
the
hum
an
capa
bi
li
t
y
to
un
der
s
ta
nd
the
la
nguag
e
.
NLP
is
the
si
gn
i
ficant
area
in
w
hi
ch
by
analy
zi
ng
data,
the
syst
e
m
ca
n
extract
the
i
nfor
m
at
ion
fro
m
the
con
te
xts
and
prov
i
des
the
input
inf
or
m
at
ion
in
seve
ral
wa
ys.
Ba
sic
al
ly
,
it
is
a
relat
ionship
am
ong
hum
an
la
nguage
an
d
com
pu
t
er
m
achine,
i.
e.
NL
P
op
e
rates
m
achine
unde
rstan
din
g,
a
naly
sis,
ge
ner
at
io
n
or
m
anip
ulati
on
of
natu
ral
la
ngua
ge.
H
ow
e
ve
r,
natu
ra
l
la
nguag
e
ref
e
r
s
to
analy
sis
of
te
xt
as
wel
l
as
aud
i
ble
s
peech,
w
her
ea
s
m
achine
cap
tures
or
rec
ognizes
the
m
eaning
of
input
w
ords
in
te
rm
s
of
struct
ur
e
d
ou
t
pu
t.
N
LP
is
ba
sic
el
em
ent
of
a
rtific
ia
l
intel
li
gen
ce
(AI).
The
i
niti
al
go
a
l
of
N
LP
is
to
pro
vid
e
s
uch
ty
pe
of
i
nterac
ti
on
s
o
that
nonpr
ogram
m
ers
can
pr
oduce
us
ef
ul
inf
or
m
at
ion
f
r
om
co
m
pu
te
r
m
achine.
S
uc
h
ty
pe
of
c
omm
un
ic
at
io
n
was
popula
rized
in
the
m
ov
ie
“
A
Sp
ace
od
ysse
y”
i
n
1968.
The
NL
P
al
so
has
the
abili
ty
to
m
ake
insig
hts
f
rom
inform
ation
co
ntained
in
m
ai
ls,
vid
e
o fil
es an
d othe
r un
st
ru
ct
ured
conten
t
[1
-
4]
.
M.
Ma
xson,
s
ai
d
that,
i
n
fut
ur
e
m
os
t
of
t
he
us
ef
ul
in
f
orm
ation
will
be
in
unstr
uct
ur
e
d
form
.
The
f
utu
re
Bi
gDat
a
will
be
t
he
com
bin
at
ion
of
bo
t
h
struc
ture
d
and
un
st
ru
ct
ur
e
d
data
and
util
iz
ing
inh
ere
nt
data
patte
r
ns
that
integrate
from
data
i
tse
lf
and
not
from
po
li
ce
i
m
po
sed
on
data
-
set
s
by
hu
m
ans.
It
ha
s
been
fr
e
qu
e
ntly
no
t
ed
that
NLP
pred
om
inantly
i
s
util
iz
ed
to
analy
ze,
retriev
e
and
s
umm
arize
the
per
ti
ne
nt
data
from
la
rg
e sets of
d
at
a avail
a
ble. An
ex
plor
at
ion
of N
L
P co
nce
pt w
as int
r
oduce
d
in 19
5
0
w
hen
T
ur
i
ng
-
te
st on
com
pu
te
r
m
ac
hin
e
a
nd
intel
li
gen
ce
was
i
ntr
oduce
d
[
5].
It
was
a
ble
to
ex
hib
it
intel
li
gent
beh
a
vior
sim
i
la
r
to,
or
no
n
-
diff
e
re
ntiable
f
r
om
,
that
of
a
pe
r
so
n.
NLP
nee
d
a
c
om
bin
at
ion
of
verbal
and
com
pu
ta
ti
on
al
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
A p
ri
or
c
as
e
stud
y
o
f
natur
al
langua
ge pr
oc
essing o
n dif
fe
rent
dom
ain
(
Shru
t
hi J
.
)
4929
knowle
dge.
T
his
can
be
do
ne
for
seve
ral
la
ng
ua
ges
f
or
e.g
.
En
glish;
sever
al
chall
eng
e
s
incu
rr
e
d
wh
e
n
the in
form
at
io
n
e
xtracti
on c
onta
in
par
a
grap
h phrasin
g, m
e
ta
phor
s
, id
i
oms an
d rh
et
or
ic
e
tc
. [6]
.
Lan
gu
a
ge
dis
a
m
big
uatio
n
i
s
the
art
t
o
par
se
an
d
e
xtract
the
in
f
orm
at
ion
in
N
L
P.
A
naly
ze
sentence
-
pa
rsi
ng,
w
ords
ide
nt
ific
at
ion
us
in
g
le
xicon,
Nam
ed
entit
y
recog
niti
on
(
NER
)
s
o
on
an
d
ext
ra
ct
ion
are
the
series
of
operati
ons
to
be
perform
e
d
in
NL
P.
Dif
fer
e
nt
ap
proac
hes
[7
]
e
xam
ple:
NER,
wr
a
ppers
i
n
web
-
c
orpus,
ta
gg
i
ng
of
pa
rts
of
s
peec
h
ha
s
bee
n
e
xplo
r
ed.
The
inte
ntion
of
i
nfor
m
at
ion
e
xtracti
on
is
t
o
extract t
he
i
nfo
rm
ation
fo
r
pre
-
determ
ined blocks f
or a s
pec
ific
f
ram
e [8
]
. Gene
ric m
od
el
s for
NLP ex
pl
oiti
ng
word
j
oi
nt
prob
a
bili
ty
and
it
s
ta
g
or
la
bel,
enh
a
ncin
g
to
Hidde
n
-
Ma
r
kov
ap
proac
h
and
trig
ram
mar
ko
v
appr
oach
has
be
en
intr
oduc
ed
in
[
9].
It
al
s
o
def
i
nes
us
age
of
ps
e
udo
ra
ndom
wo
r
ds
a
nd
their
f
unct
ion
i
ng
f
or
NLP
of
unide
nt
ifie
d
words
for
sm
a
ll
dataset.
In
[
10]
Lapata
et
.al
hav
e
il
l
us
trat
ed
a
bout
web
-
based
a
pproa
c
h
for
NL
P
that
pro
vid
e
d
th
e
kn
ow
le
dg
e
of
iso
la
ti
on
of
com
po
un
d
no
un
s
,
di
so
r
de
rly
arr
a
nge
m
ent
of
a
dj
ec
ti
ves,
web
inter
pola
ti
on
an
d
m
ass
counts
t
o
m
axim
iz
e
the
flexi
bili
ty
.
Thus,
N
LP
fall
s
i
nto
the
do
m
ai
n
of
AI
with
the aim
o
f
un
de
rstan
ding a
nd
m
aking
m
eaningf
ul info
rm
ati
on in
t
he hu
m
an
la
ng
uag
e
[1
1]
.
Diff
e
re
nt
te
rm
ino
lo
gies
a
re
util
iz
ed
in
N
L
processi
ng
suc
h
as
1)
M
orpholo
gy:
-
T
his
is
us
e
d
to
const
ru
ct
a
m
eanin
gful
wor
d
f
r
om
a
pr
im
it
ive
sentence
,
2)
Synta
x:
-
util
iz
ed
f
or
w
ord
arr
a
ngem
ent
wh
ic
h
resu
lt
s
a
m
ean
ingful
sente
nc
e,
3)
Sem
ant
ic
s:
-
con
ce
rn
e
d
with
m
eaningfu
l
w
ord
an
d
c
reate
the
m
eaningfu
l
sentence
by
c
om
bin
ing
th
os
e
words
a
nd
ph
r
ases,
4)
Pr
a
gm
at
ic
s:
-
util
iz
ed
to
know
t
he
m
eaning
of
se
ntence
for
di
ff
e
ren
t
s
it
uations
a
nd
how
the
un
de
rstan
ding
of
s
entence
is
e
ffec
te
d
[
12
]
.
Th
e
re
m
ai
nin
g
pa
rt
of
the
stud
y
is
organ
iz
e
d
as;
sect
ion
-
2
desc
ribe
s
the
gen
e
ric
proces
s
of
NLP
wh
ic
h
diag
ram
m
at
ic
ally
rep
resen
ts
the
pr
ocedur
al
ste
ps
of
la
ngua
ge
processi
ng.
Sect
ion
-
3
disc
us
ses
sco
pe
a
nd
a
pp
li
cat
io
ns
of
NL
P
f
ollo
w
ed
by
diff
e
re
nt
te
chni
qu
es
a
nd
ap
pr
oach
e
s
util
iz
ed
in
NLP
in
sect
ion
-
4.
Sect
ion
-
1.2
highli
ghts
the
existi
ng
res
earch
work
t
ow
a
rds
NLP
im
ple
m
e
ntati
on
f
ollo
w
ed
by
crit
ic
al
chall
enges
in
NLP
in
Sect
i
on
5.
Finall
y,
sect
ion
-
6
pro
vid
es t
he
c
oncl
us
i
on of the
r
esea
rch stu
dy.
This
sect
ion
discu
sses
s
om
e
of
the
e
xisti
ng
te
chn
iq
ues
of
i
m
ple
m
enting
NLP
.
Alm
ada
et
al
.
[1
3],
pr
ese
nts
a
n
ide
ntifyi
ng
te
c
hn
i
cal
know
le
dge
(TK)
in
s
oft
wa
re
de
velo
pm
ent.
It
help
s
in
th
e
hiring
of
s
oft
ware
eng
i
neer
i
ng
po
sit
ion
s
a
nd
ta
le
nt
m
anag
em
ent.
To
de
velo
p
this
NLP
te
c
h
niq
ue
is
us
e
d
a
nd
te
xt
m
ining
(
TM)
is
util
iz
ed
to
analy
ze
form
le
s
s
te
xt
in
res
ume
s
an
d
cu
rr
ic
ulu
m
.
By
i
m
p
leme
nting
K
P
G
ENER
ATO
R
,
sp
eci
fi
c
data
on
res
ume
s
is
analy
zed
and
sig
nifica
ntly
red
uce
t
he
tim
e
of
w
ork.
Pela
yo
et
al
.
[14]
pr
ese
nt
ed
an
enh
a
ncin
g
hea
ri
ng
im
paired
-
stud
e
nt
read
i
ng
s
kill
s
in
col
le
ge
le
vel.
T
he
m
ai
n
obj
ect
ive
of
this
sub
j
ect
is
fo
c
us
e
d
on
on
ly
deaf
stude
nt
s.
The
an
al
ysi
s
of
te
xt,
im
ages
and
oth
e
r
inf
or
m
at
ion
with
res
pect
to
c
on
te
xt
NLP
act
as
m
ai
n
r
ole.
Her
e,
the
us
er
ca
n
im
agine
t
he
in
for
m
at
ion
by
gra
dual
ly
increasin
g
the
widt
h
of
words
and can
im
pr
ove
unde
rstan
din
g.
Kh
al
id
et
al
.
[15],
pr
ese
nt
ed
ide
ntifyi
ng
the
m
ulti
-
wo
r
d
Urdu
f
rom
a
nu
m
ber
of
co
rpo
ra.
The
im
pr
ov
e
m
ent
of
this
schem
e
C
-
value
m
et
ho
d
an
d
NL
P
is
util
iz
ed
al
ong
w
it
h
Urdu
Wor
d
Net
.
The
outc
om
es
of
this
m
et
ho
d
com
pr
ise
s
of
735
synset
s
co
ntainin
g
only
m
ult
i
-
wor
d
-
U
rdu
te
rm
s
from
Urdu
-
Wor
dN
et
. J
i
m
enez et al
.
[
16
]
h
a
ve
pr
opose
d
to
pr
e
dict t
he
data o
r
searchi
ng
in
form
at
ion
u
si
ng
Bi
g Data
pr
e
dicti
ve
a
naly
ti
cs
in
so
ci
al
m
edia.
The
re
s
ult
disc
ov
e
re
d
and
analy
ze
d
present sta
tus o
f
the r
esearc
h
t
ha
t
has
been
urba
nized
so
far
by
edu
c
at
ion
al
bac
kgr
ound.
Ela
hi
nia
et
al
.
[1
7],
eva
luate
d
a
pr
e
dic
ti
on
of
com
plex
an
d
m
ul
ti
di
m
ension
al
healt
h
ca
re
pro
blem
us
ing
NLP
te
c
hn
i
que.
The
m
ai
n
obj
ect
iv
e
of
this
stud
y,
first
an
al
yz
ed
the
m
os
t
com
p
reh
e
ns
ive
el
em
ent
of
he
al
thca
re
data
beca
use
of
m
any
-
side
d
nat
ur
e
of
the
pro
blem
.
The
ou
t
pu
t
sh
ows
t
he
poss
ibil
it
y
of
patie
nt
-
non
a
dh
e
re
nc
e
by
pe
rs
on
ge
ner
at
e
d
in
put
-
word
(
PGI
W)
and
com
bin
at
ion
of
m
achines.
Slat
er
et
al
.
[
18
]
,
de
velo
pe
d
a
n
e
ff
ect
i
ve
m
easur
e
of
stud
e
nt
a
ppoi
ntm
ent
in
tuit
ion
or
cl
ass
.
The
te
ch
nique
NLP
util
iz
ed
to
identify
the
dif
fer
e
nt
li
nguisti
c
feature
s
su
ch
as
m
at
hem
atics
pr
oble
m
s
on
st
ud
e
nts
-
a
ff
ect
ive
sta
te
s
durin
g
w
ork
in
a
m
at
he
m
at
ic
-
tutor
.
The
outc
om
e
is
m
easur
ed
with
a
li
ng
uisti
c
featur
e,
a
nd
stu
de
nt
pro
blem
s
on
on
li
ne
tut
or
are
evaluate
d.
Ca
la
podesc
u
et
al
.
[1
9]
present
e
d
a
sem
i
-
autom
a
ti
c
de
-
ide
ntific
at
ion
of
hosp
it
al
discha
rg
e
s
umm
aries.
The
pro
po
se
d
syst
e
m
is
achieve
d
wit
h
sign
ific
a
nt
im
p
rovem
ents
fo
r
the
an
no
ta
ti
on
of
t
he
doc
um
e
nts
in
qu
a
ntit
at
ive,
qual
it
at
ive
and
hom
og
en
eous
resu
lt
s.
The
w
ork
of
Br
on
ie
c
ki
et
al
.
[20]
ha
ve
fo
c
us
e
d
on
the
upc
om
in
g
a
rr
ivi
ng
f
orm
of
data
a
nd
to
m
ake
internati
onal
pract
it
ion
ers
grow.
T
he
prese
nting
new
data
w
ork
on
la
rge
quest
io
n
te
xt
an
d
al
s
o
w
orked
on
sp
eci
fic a
pproa
ches. The
outc
om
es are s
ta
ble on hum
an
-
de
rive
d
co
ding a
nd
com
plete
on c
orrelat
e/
pr
e
dict an
d
achieve t
he dat
a pro
gr
am
.
In
the
st
u
dy
of
Zit
nik
et
al
.
[
21]
ha
ve
c
oncen
trat
ed
on
the
va
rio
us
ty
pes
of
ap
plica
ti
on
s
a
nd
to
ols
tha
t
is
require
d
for
m
achine
le
arn
in
g
i
n
Natu
ral
la
ngua
ge
proces
s.
A
new
arr
i
ved
ap
plica
ti
on
cal
le
d
t
oo
l
kit
pro
vid
e
an
e
nd
to
en
d
te
xt
inv
e
sti
gation
by
the
us
ing
of
JavaScr
i
pt
te
chn
i
qu
e
base
d
f
un
ct
io
n.
T
he
s
ta
te
of
the
arts
a
re
co
m
es
in
the
te
r
m
s
of
the
res
ul
t.
The
work
of
Ba
by
et
al
.
[
22
]
hav
e
f
ocused
on
hose
c
on
t
rol
app
li
cat
io
n
in
wh
ic
h
the
us
e
r
can
inter
face
with
we
b
a
ppli
cat
ion
or
a
ne
wly
arr
iv
ed
c
ha
tbo
t
to
ols
it
s
con
t
ro
l
on
al
l
ho
us
e
el
ect
rical
equ
ip
m
ent.
The
pr
es
enting
c
hatb
ot
te
chn
iq
ue
fo
ll
ow
the
sta
te
m
ent
of
te
xt
in
for
m
at
ion
that’s
te
xt
by
us
er
i
n
the
f
orm
of
com
m
and
a
nd
t
hen
co
ntr
ol
the
al
l
-
el
ect
ric
equ
i
pm
e
nt
at
house
by
us
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol. 10,
No.
5,
Oct
ob
e
r 2
020
:
49
28
-
49
36
4930
the
m
otion
sen
so
rs
.
T
he
st
udy
o
f
Ka
us
hi
k
e
t
al
.
[23]
hav
e
fo
c
us
e
d
on
iss
ues
of
NL
P
in
wh
ic
h
they
dis
cusse
d
on
dee
p
le
ar
nin
g
w
hich
s
olve
s
the
pro
ble
m
of
N
-
la
ng
uag
e
proces
sing.
By
the
ef
fecti
ve
dee
p
l
earn
i
ng
the
secret
iss
ue
s
li
ke
outp
ut
-
vecto
r
an
d
in
put
-
word
a
re
re
m
ov
ed.
By
the
add
i
ng
of
Cr
os
s
-
Lan
guage
fu
t
ur
e
work
is
done
.T
he
stu
dy of Ah
m
ed
et
al.
[24] h
ave c
oncent
ra
te
d
on poetry a
naly
zer on N
-
l
angua
ge
pr
oce
ssing
wh
ic
h
us
e
d
dif
fer
e
nt
ty
pes
of
la
nguag
e
a
nd
i
ts
con
side
r
the
com
pu
te
r
-
base
d
ap
proac
h.
Th
is
m
et
ho
d
cons
iders
al
l
ty
pes
of
li
te
rar
y
te
ste
d
be
fore
prese
ntin
g
the
final
re
s
ult
is
ob
ta
i
ned.Th
e
w
ork
of
po
le
et
al
.
[25
]
hav
e
fo
c
us
e
d
on
the
new
pro
posed
appr
oach
for
th
e
instru
ct
io
n
of
groupin
g
do
c
um
ents’
in
w
hic
h
searc
h
re
su
lt
s
are
in
the
te
rm
s
of
bette
r
way
-
group
f
uzzy
ap
pa
ratus
f
or
ac
cu
rate
com
pilat
io
n
by
the
us
in
g
of
weig
hted
m
et
hod.
In
the
stu
dy
of
Ba
rtero
et
al
.
[26]
descr
i
be
d
an
om
al
y
identific
at
ion
pr
ob
le
m
.
To
res
olv
e
this
issue
[26]
introd
uced
a
Wor
d
-
em
bed
di
ng
m
et
ho
d
bas
ed
on
Goo
gle’
s
Word2
vec
al
gorithm
.
The
resu
lt
s
il
lustrate
d
90%
accuracy a
nd e
ff
ic
ie
nt c
om
pu
ta
ti
on
.
Suhaim
in
et
al
.
[27]
de
scri
bed
a
chall
e
nge
wh
ic
h
occ
ur
s
i
n
sa
rcas
m
detection
i
nto
t
he
te
xts.
The
ou
tc
om
es
show
t
hat
a
gro
up
of
pro
s
od
ic
,
synta
ct
ic
an
d
pra
gm
ati
c
struct
ur
es
de
velo
ped
an
e
ff
ect
iv
e
perform
ance
w
it
h
F
-
m
easur
e
resu
lt
of
0.8
52.
The
stu
dy
car
r
ie
d
out
by
Ala
m
i
et
a
l.
[2
8]
ha
s
pr
ese
nte
d
a
novel
Sem
i
-
autom
at
e
d
m
et
ho
do
l
ogy
to
dev
el
op
a
sequ
e
ntial
struct
ur
e
f
r
om
us
er
necessit
ie
s
wr
it
te
n
in
Ar
a
bic.
The
re
su
lt
sho
ws
that
propos
ed
ap
proac
h
prov
i
des
a
n
ef
fec
ti
ve
U
ni
fied
M
od
el
in
g
La
ngua
ge
(
UML)
dia
gr
am
s
from
Ar
abic
use
r
re
quirem
ents.
Zh
ua
ng
et
al
.
[29]
prese
nt
ed
a
Str
ok
e
L
evel
ap
proac
h
to
infl
uen
ce
C
hin
ese
stroke
f
or
le
a
r
ning
the
Chi
ne
se
wor
d
re
pr
es
entat
ion
.
The
e
xp
e
rim
ental
ou
tc
om
es
sh
ow
t
hat
presente
d
m
et
ho
d
is
s
uperio
r
for
Chinese
-
Lan
gu
age
processin
g.
To
re
duce
thi
s
ty
pe
of
pro
ble
m
the
Des
hmukh
et
al
.
[
30
]
has
dev
el
op
e
d
V
oice
-
base
d
a
ppli
cat
ion
to
the
us
ers
.
T
he
outc
om
es
disp
la
y
the
pro
po
se
d
syst
e
m
that
prov
i
de
s
correct
answe
r t
o
the
us
e
r.
It
offe
rs fast p
r
oce
ssin
g an
d
m
axim
u
m
r
espo
nse
capab
il
it
y t
o
th
e u
se
rs.
2.
GENERI
C P
ROCESS
OF
NLP
Gen
e
rall
y,
the
re
ar
e
fi
ve
st
eps
i
n
NLP
i.
e.,
M
orpholo
gi
cal
,
Lexical
analy
sis,
Synt
ax
a
naly
sis,
Pr
og
ram
m
at
ic
s
and
Disc
ours
e.
The
proces
s
of
natu
ral
la
ngua
ge
ne
eds
knowle
dge
of
ho
w
utte
ra
nc
e
ar
e
create
d,
how
they
turn
int
o
cl
auses
and
sentences
.
A
ddit
ion
al
ly
,
to
eff
ect
ively
under
sta
nd
a
gr
oup
of
sentences
in
t
he
g
ive
n
c
onte
xt
, it re
qu
ire
s a
hi
gh
le
vel kn
owle
dg
e.
S
uch lev
el
s ar
e
descr
i
be
d
as
foll
ows;
a.
Mor
phologica
l:
It
deals
with
m
or
pholog
ic
al
-
str
uctur
e
of
w
ords,
e
.g.
prefi
x,
s
uffix,
w
ord
root,
a
nd
i
nf
ix
es
.
Ba
sic
al
ly
,
m
or
ph
em
e
is
the
basic
un
it
of
wr
it
te
n
w
ord
wh
ic
h
pro
vid
e
s
the
i
nfor
m
ation
a
bout
w
or
d
form
ation
.
b.
Lexical
A
naly
sis:
In
this
sta
ge,
ea
ch
w
ord
is
analy
zed
w
it
h
their
c
om
po
ne
nt
a
nd
othe
r
non
-
w
or
ds
l
ik
e
punctuati
ons
,
acronym
s
and
abbrev
ia
ti
ons
are
separ
at
e
d
from
wo
r
ds.
This
analy
ti
cal
pr
ocess
is
respo
ns
ible
for
identify
in
g
a
nd
a
naly
zes
the
word
str
uctu
re
and
se
par
at
in
g
the
w
hole
te
xt
into
par
a
gr
a
ph
s,
set
o
f
se
ntence
s and
wor
ds
.
c.
Synta
x
A
naly
sis:
The
pr
im
ar
y
pu
r
pose
is
t
o
fin
d
that
a
words
strin
g
is
well
arr
an
ged
and
di
vid
e
it
int
o
a
struct
ur
al
f
or
m
at
wh
ic
h
r
ep
resen
ts
t
he
rela
ti
on
be
twee
n
the
disti
nct
wor
ds
.
It
deals
with
w
ord
st
ru
ct
ure
and
v
al
idit
y
of
sen
te
nces
.
A
s
ynta
ct
ic
analy
z
er
p
er
f
or
m
s
thi
s
byex
plo
it
ing
a
word
le
xic
on
and
synta
x
r
ule
s
(i.e., G
ram
m
ar)
.
d.
Sem
antic
A
naly
sis:
I
t pro
vid
e
s ex
act
m
eanin
g
f
r
om
the g
ive
n
te
xt.
T
he
te
xt
is v
e
rified
for
m
eaningfu
l
nes
s.
The
a
naly
zer
r
ejects
the
se
nte
nce
li
ke
‘I
ll
ega
l
-
Law”.
The
se
ntence
st
ru
ct
ure
create
d
by
synta
x
-
analy
zer
i
s
assigne
d
m
eaningf
uln
ess
.
e.
Pr
a
gm
atics:
In
a
sp
eci
fic
c
onte
xt,
it
de
al
s
with
the
sent
ences
wh
e
re
t
he
se
ntence
st
ru
ct
ur
e
repres
ent
s
the
act
ual
m
ea
ning
of
that
se
ntence
f
or
exa
m
ple
:
“cl
os
e
t
he
wi
ndow
?
”
It
has
bee
n
int
erprete
d
as
a
ppeal
rather t
ha
n ord
er [3
1].
f.
Disco
ur
s
e:
Th
e
m
eaning
of
sp
eci
fic
sent
ence
m
ay
depends
upon
se
ntence
prece
di
ng
an
d
m
igh
t
be
influ
e
nce
t
he
sign
ific
a
nce
of
that
sente
nce.
For
e
xam
ple:
“He
wan
te
d
i
t”
wh
ic
h
de
pe
nd
s
on
e
xisti
ng
disco
urse c
on
t
ext.
3.
ESSENTI
ALS O
F
N
LP
It
is
well
kn
own
t
hat,
NLP
is
a
br
a
nc
h
of
AI
w
hich
has
seve
ral
sig
nif
ic
ant
associat
ion
s
bet
wee
n
com
pu
te
r
m
ac
hin
e
a
nd
pe
rs
on
interact
io
n.
I
n
this
dig
it
al
world
,
NL
P
play
s
a
sign
ific
a
nt
r
ole
w
hich
br
i
dg
i
ng
the
com
m
un
ic
at
ion
ga
p
am
on
g
hu
m
an
an
d
dig
it
al
wo
rl
d.
Fu
rt
her
m
or
e,
this
sect
ion
pro
vid
es
sc
ope
and
app
li
cat
io
ns
of N
LP
whic
h
is
util
iz
ed
in t
he day
s to
co
m
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
A p
ri
or
c
as
e
stud
y
o
f
natur
al
langua
ge pr
oc
essing o
n dif
fe
rent
dom
ain
(
Shru
t
hi J
.
)
4931
3.1.
Scop
e
of
NLP
3.1.1
.
M
achin
e
tr
an
sla
tio
n
As
t
he
global
i
nfor
m
at
ion
is
on
li
ne
,
t
he
acc
essibil
it
y
of
in
form
ation
is
be
com
ing
hi
gh
l
y
increasi
ng.
The
m
ai
n
chall
eng
e
is
to
pro
vi
de
acce
ss
for
globali
nfor
m
at
ion
t
o
e
ver
y
pe
rson
with
unde
rstan
dab
le
la
nguage
translat
ion.
Mo
dern
com
pan
ie
s,
e.g.,
“D
uoli
ngo,”
recr
uit
a
num
ber
of
em
pl
oyees
to
co
ntri
bu
te
,
by
con
c
urrin
g
translat
ion
w
ork
with
e
ducat
ing
ne
w
la
ng
ua
ge.
B
ut
di
gital
translat
ion
pro
vid
es
m
or
e
scal
able
al
te
r
native
inf
or
m
at
ion
to
co
rr
es
pondin
g
global
in
for
m
at
ion
.
G
oogl
e
is
the
i
nnovat
ive
c
om
pany
that
has
m
a
chine
translat
ion
w
hi
ch
tra
ns
la
te
s
th
e
inf
or
m
at
ion
base
d
on
t
he
pro
pr
ie
ta
ry
sta
ti
sti
cal
inform
ati
on.
T
he
c
halle
ng
e
of
m
achine
tra
ns
l
at
ion
te
ch
ni
que
is
not
tra
nsl
at
ing
the
w
ords
bu
t
unde
rs
ta
nd
s
an
d
st
ores
the
m
eanin
g
of
the se
ntences t
o offer
an act
ua
l t
ran
sla
ti
on.
3.1.2
.
Auto
ma
tic
summ
ariz
at
ion
Durin
g
acce
s
sing
of
sp
eci
fic
and
sig
ni
ficant
inf
or
m
at
ion
from
t
he
exten
sive
know
le
dg
e
,
an
in
form
at
ion
overlo
adi
ng
is
a
com
plex
c
ha
ll
eng
e
to
retrie
ve
s
uch
in
form
at
ion
.
A
uto
m
atic
su
m
m
ariza
tio
n
is
no
t
only
releva
nt
to
s
umm
ariz
e
the
m
eaning
of
t
he
docum
ent,
bu
t
al
s
o
t
o
unde
rstan
d
the
em
otion
al
m
eanin
g
inside
the
inf
or
m
at
ion
,
e.g.,
wh
il
e
colle
ct
ing
or
acce
ssi
ng
in
form
at
io
n
from
the
so
ci
al
sit
es.
Thi
s
NLP
app
li
cat
io
n
is
be
com
ing
m
or
e u
sef
ul as
a
de
m
and
able m
ark
et
asset
.
3.1.3
.
I
nform
ati
on
ex
tra
c
tio
n
Durin
g
the s
um
m
arization
of
so
m
e k
ey
asp
ect
s o
f
s
pecific kin
d
of sto
ries,
w
e u
se
patte
rns ex
tract
io
n.
Fo
r
exam
ple,
the
sta
ti
sti
cs
on
te
rror
ist
at
ta
ck
s
can
be
s
umm
arized
by
seei
ng
place
,
date
,
a
n
id
of
pe
r
petr
at
or
s,
injur
ie
s,
a
nd
de
at
hs
,
et
c.,
are
exp
r
essed
,
an
d
there
fore
ex
plorin
g
out
that
info
rm
at
ion
and
ex
plore
it
into
a predeterm
ined
te
m
plate
.
3.1.4
.
Ques
tio
n
answ
eri
ng
At
pr
ese
nt,
Q
A
is
beco
m
ing
m
or
e
popu
l
ar
for
the
hum
ans,
an
d
tha
nk
s
t
o
the
ap
plica
ti
on
s
li
ke;
OK
-
G
oog
le
,
S
iri,
an
d
chat
-
boxe
s.
T
hese
ki
nd
s
of
ap
plica
ti
on
s
a
re
ca
pa
ble
to
ans
we
r
the
hu
m
an
re
quest
s
.
It
can
be
util
iz
ed
as
a
te
xt
interface
or
li
ke
a
li
ng
uisti
c
dialog
syst
em
.
More
ov
e
r,
this
app
li
cat
ion
re
m
ai
ns
m
or
e
chall
engi
ng
es
pecial
ly
fo
r
sea
rch
e
ngines
,
a
nd
it
i
s
the
on
e
of
the
sig
nificant
app
li
cat
io
n
of
NLP
researc
h fiel
d.
3.2
.
NLP
techniq
u
es
The NLP
is
use
d
in
d
i
ff
e
ren
t
a tec
h
ni
que wh
ic
h
is
descr
i
be
d belo
w:
3.2.1.
N
amed
-
entit
y
rec
ognition
(
NER
)
The
NER
is
a
ty
pe
of
inf
or
m
at
ion
/data
extr
act
ion
that
go
a
l
to
reco
gniz
e
and
determ
ine
ph
ra
ses
or
words
in
te
xt
form
at
into
pre
-
de
fine
d
cl
asses
li
ke
as
the
nam
es
of
places,
perso
ns
,
e
xpressi
ons,
et
c.
[
32
].
The
nam
ed
en
ti
ty
(N
E)
is
an
orde
r
of
w
ords
t
hat
des
crib
e
real
-
world
obj
ect
su
c
h
a
s
Apple
In
c
orp
orat
io
n,
Ca
li
fo
r
nia,
et
c.
In
data
ext
racti
on
,
t
he
N
ER
is
an
esse
ntial
ta
sk
.
T
he
NER
con
ta
in
s
thr
ee
un
i
ver
sal
ly
acc
epted
ty
pes
are;
Plac
es,
Pe
rs
on
a
nd
O
r
gan
iz
at
ion.
NER
doe
sn
’t
ge
ner
at
e
tem
plate
s,
and
it
is
not
a
n
eve
nt
recog
niti
on
[
33
]
. Th
e
re ar
e
s
om
e
m
et
ho
dolo
gies
pr
e
se
nted
wh
ic
h
a
re
us
ed
in NER
proces
ses are:
a.
Rule
-
ba
sed
M
et
hodo
l
og
y:
T
he
Rule
-
base
d
appr
oach
w
ork
as
a
com
bin
at
ion
of
r
ules
ei
ther
a
uto
m
at
ic
a
ll
y
or
m
anu
al
ly
de
fine
d.
A
r
ule
con
ta
in
s
act
ion
and
a
patte
r
n.
The
Am
azi
gh
-
nam
ed
entit
y
reco
gnit
ion
(
A
ER)
base
d
on
s
ym
bo
li
c
m
et
ho
d
us
es
li
ng
uisti
c
r
ules
m
anu
al
ly
to
gen
e
rate
t
he
im
pr
ovem
e
nt
of
gazett
eer
s
by
us
in
g
intr
oduc
ed
ap
proac
h
[
34
]
.
T
he
NER
syst
e
m
is
us
ed
to
i
m
pr
ove
the
new
sta
ndard
Ar
a
bic
with
the h
el
p of t
he R
ule
-
ba
sed
m
e
thod
[35].
b.
Stat
ist
ic
al
Lear
ning
m
et
ho
d:
The
m
uch
Stat
ist
ic
al
Learn
in
g
m
et
ho
d
is
ba
sed
on
N
ER
al
gorithm
s,
and
it
treat
s
the
w
ork
as
a
Se
quen
ce
-
la
belin
g
P
r
ob
le
m
(S
LP).
The
SL
P
is
a
com
m
on
m
achine
le
arn
i
ng
(
ML)
pro
blem
and
has
bee
n
helped
to
c
reate
dif
fer
e
nt
NL
P
w
orks
al
so
co
ntainin
g
c
hunkin
g,
NE
R
,
and
par
t
-
of
-
s
pe
ech
ta
ggin
g.
The
Stat
ist
ic
al
Learn
i
ng
a
pp
ro
ac
h
incl
ud
es
three
m
od
ule
s
li
ke
Hiddde
n
Ma
rkov m
od
el
s,
Ma
xim
u
m
en
tr
op
y M
a
rko
v m
od
el
s and
c
onditi
onal
r
a
ndom
fiel
ds
.
3.2.2.
Speech
recogni
tio
n m
eth
od
The
NLP
is
a
m
et
ho
d
t
hat
ca
n
m
ini
m
iz
e
the
distance
bet
ween
m
achine
an
d
hu
m
an
be
ing
.
It
m
akes
hu
m
an
to
com
m
un
ic
at
e
with
the
de
vice
wit
hout
any
c
om
plexity
.
The
S
pe
ech
Re
co
gnit
ion
has
t
he
cap
abili
t
y
of
a
program
or
m
achine
to
detect
phrase
s
or
w
ords
i
n
a
ve
r
b
al
la
ngua
ge
and
c
ha
ng
e
them
to
a
m
achine
-
u
nd
erstan
dab
le
form
at
.
It
is
a
t
echnolo
gy
w
hi
ch
al
lows
hu
m
an
com
m
un
i
cat
ion
with
de
vices.
These
are
benefici
al
in
every
day
li
fe
becau
se
a
m
achine
can
unde
rstan
d
by
voic
e.
Th
e
NLP
pe
rform
s
ver
y
eff
ic
ie
ntly
to
create
com
pu
te
r
-
inter
faces
that
m
ake
m
or
e
com
fo
rt
able
to
us
e
f
or
hu
m
ans.
The
m
od
er
n
NLP
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol. 10,
No.
5,
Oct
ob
e
r 2
020
:
49
28
-
49
36
4932
m
et
ho
d
al
lo
ws
the
us
e
of
N
L
t
o
m
anifest
pro
gr
am
m
ing
i
deas
[
36
]
.
T
he
Figure
1
sho
ws
the
w
ho
le
sp
eec
h
recog
niti
on
w
orki
ng
a
nd
is
distribu
te
d
in
f
our
sta
ges.
The
featur
e
ext
racti
on
(FE)
proces
s
is
ap
plied
by
us
in
g
Me
l
fr
eq
ue
ncy
cepstral
coe
ffi
ci
ent
(MFCC
)
in
w
hich
vo
ic
e
featur
e
s
are
extracte
d
f
or
e
ver
y
vo
ic
e
sa
m
ples.
Af
te
r
t
hat,
al
l
fu
nctio
ns
a
re
spe
ci
fied
to
Patt
e
rn
trai
ning
w
hi
ch
is
trai
ne
d
by
hid
de
n
m
ark
ov
m
od
el
(
HM
M)
to
gen
e
rate
HM
M
fo
r
eac
h
w
ord.
At
la
st
t
he
Vite
r
bi
dec
od
i
ng
is
us
e
d
to
choose
th
e
on
e
wit
h
m
axim
u
m
pro
bab
il
it
y wh
i
ch
is
nothin
g b
ut r
ec
ognize
d word [
37
].
Figure
1
.
S
pee
ch
-
r
eco
gnit
ion
s
yst
e
m
The Speec
h
-
re
cogniti
on syst
em
co
ntains
f
ollow
i
ng tech
niques
:
a.
Dynam
ic
Ti
m
e
W
a
r
ping
(DT
W)
:
The
DT
W
m
et
ho
d
is
a
Dynam
ic
Pr
ogra
m
m
ing
m
et
ho
d
wh
er
e
the
w
hole
issue
is
distrib
uted
into
a
sm
al
l
no
.
of
ste
ps
and
eve
ry
ste
p
nee
ding
a
decisi
on
to
be
create
d
base
d
on
the
L
ocal
-
distance
m
easur
es.
The
w
ho
le
dec
isi
on
is
de
pe
ndin
g
on
sm
al
l
decisi
ons.
The
co
nque
rin
g
a
nd
div
ide
m
et
ho
dolo
gy sti
m
ulate
the
DT
W.
T
he
D
T
W
c
ould
f
ind
t
he
m
ini
m
um
p
at
h
distanc
e w
it
h
t
he help
o
f
m
at
rix,
and als
o
ca
n red
uce t
he
co
m
pu
ta
ti
on
a
m
ou
nt
[
38
]
.
b.
Hidden
Ma
r
ko
v
M
od
el
s
(
H
MM
):
In
HMM
sta
ti
sti
cal
design
the
syst
e
m
bein
g
desi
gn
e
d
us
es
Ma
rko
v
proce
dure
with
unknown
par
a
m
et
ers.
The
ai
m
is
to
def
ine
the
unknow
n
pa
ram
et
ers
fr
om
the
ob
ser
va
ble
inf
or
m
at
ion
.
I
n
HMM,
the
sta
te
is
no
t
di
rectl
y
visible,
bu
t
t
he
v
aria
bl
es
aff
ect
e
d
by
the
sta
te
w
hi
ch
are
visible.
c.
Vecto
r
Q
uan
ti
zat
ion
(
VQ):
The
V
Q
is
a
proce
dure
f
or
m
app
in
g
vect
ors
from
a
big
vector
s
pace
to
a
lim
it
ed
nu
m
ber
of
cl
us
te
r/r
egio
ns
in
that
sp
ace.
T
he
cl
ust
ering
/m
app
in
g
the
vecto
rs
can
be
achie
ved
by
us
in
g
a
clusteri
ng alg
or
it
hm
su
ch
as LB
G (L
ind
eB
uzo an
d Gr
ay
)
[3
9].
d.
Ergo
dic
Hi
dde
n
Ma
r
kov
Mo
dels
(
E
-
HMM):
Th
e
HMM
is
the
opti
m
al
ap
proach
f
or
dem
on
strat
i
ng
the
voic
e
si
gn
a
l
because
the
HMM
is
sig
nif
ie
d
in
te
rm
s
of
sta
te
s
w
hich
i
s
the
c
har
act
e
r
ist
ic
s
of
t
he
vo
ic
e
sign
al
a
nd
it
is
al
so
sho
wn
i
n
te
rm
s
of
sta
te
s.
The
E
-
HMM
is
us
e
d
f
or
m
odel
ing
the
vo
ic
e
sign
al
.
E
-
HM
M
can
directl
y cr
eat
e the
order
of v
e
rb
al
elem
ents in
a se
qu
e
nce
[40].
e.
Ar
ti
fici
al
Ne
ural
Netw
ork
(
A
NN):
A
n
ANN
is
a
pro
gr
am
,
wh
ic
h
trie
s
to
cop
y
t
he
a
ct
io
n
of
the
bi
olog
ic
al
act
ivit
y
of
the
hu
m
an
m
ind
.
The
ANN
ha
s
been
ef
fectu
al
with
ov
erla
pp
i
ng,
noisy
and
var
ia
ble
data
stream
s.
It
pro
vid
es
e
ff
ect
iv
e p
er
form
ance
in
vo
ic
e rec
ogniti
on
b
eca
us
e
it
delivers
a
faste
r
com
m
un
ic
at
ion
as
com
par
ed
t
o
“Te
xt
Wr
it
ing
,”
it
has
a
ha
nd
fr
ee
abili
ty
,
ben
e
fici
al
for
m
ental
ly
or
ph
ysi
cal
ly
disabled
peoples
[41
].
3.2.3.
Op
tical
char
act
er
rec
ognitio
n
(
OCR)
sy
s
tem
The
OCR
syst
e
m
per
m
i
ts
to
scan
hand
or
ty
pew
ritt
en
t
ext
an
d
cha
ng
e
the
scan
ne
d
picture
i
nto
a
com
pu
te
r
process
form
at
.
It
co
uld
be
ei
ther
i
n
a
w
ord
docum
ent
or
plain
te
xt
f
orm
.
The
trans
f
or
m
ed
do
c
um
ent
can
be
m
od
ifie
d,
or
us
e
d
in
dif
fe
ren
t
do
c
um
ents.
The
re
by,
th
e
do
c
um
ents
file
beco
m
es
edita
ble.
The
OCR
is
us
ed
wh
e
n
ren
e
wing
a
sim
il
ar
do
c
um
ent
fi
le
in
a
pag
e
as
a
do
c
um
ent
for
m
at
in
el
ect
ro
ni
c
form
ta
kin
g
e
xtra
ti
m
e.
It
is
extensively
us
e
d
in
trans
f
or
m
ing
books
a
nd
t
exts
form
at
in
to
the
dig
it
al
form
.
Hen
ce
,
it
ca
n
be
us
e
d
in
te
xt
m
ining
,
m
ac
hin
e
tran
sla
ti
on
,
c
ha
racter
-
re
cogniti
on
a
nd
el
ect
ro
nical
l
y
search
.
T
he
t
hr
ee im
portant t
ech
nique
s for
OCR
proc
essing [
42
]
are
giv
e
n belo
w
:
a.
Pr
e
-
Processi
ng:
The
Pr
e
-
proc
essing
ste
pis
to
gen
erate
data
that
are
ver
y
easy
and
eff
ic
ie
nt
for
the
OCR
to
op
e
rate
an
d
pe
rfor
m
accurate
ly
.
The
P
re
-
processin
g
sta
ge
is
to
en
hance
the
pictu
re
qual
it
y
fo
r
opti
m
al
recog
niti
on
by
the
syst
e
m
.
Ther
e
a
re
fe
w
te
chn
iq
ues
w
hich
im
ple
m
e
nt
this
proces
s
are
De
-
S
ke
w,
Bi
nar
iz
at
ion, L
ine
-
rem
ov
al
,
etc.
b.
Segm
entat
ion
:
The
ne
xt
OCR
processin
g
ste
p
is
segm
entation
,
where
the
char
act
er
picture
i
s
se
gm
ent
e
d
into
it
s
su
b
-
c
om
po
nen
t
f
or
m
.
It
is
essenti
al
because
if
on
c
e
extent
enters
into
a
separ
at
i
on
of
the
dif
fere
nt
li
nes
in t
he
c
ha
racters,
it
aff
ec
ts t
he
rec
ogniti
on
-
rate
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
A p
ri
or
c
as
e
stud
y
o
f
natur
al
langua
ge pr
oc
essing o
n dif
fe
rent
dom
ain
(
Shru
t
hi J
.
)
4933
c.
Po
st
-
Processi
ng:
T
he
P
os
t
-
processi
ng
do
eserror
-
c
orrecti
on
/dete
ct
io
n
and
gr
ou
ping.
It
re
fines
O
CR
ou
tc
om
es
with
the
help
of
s
pell
or
gr
am
m
ar
chec
k
a
nd
oth
e
r
sp
eci
fic
knowle
dge
-
s
ou
rce
com
par
iso
ns.
The res
ult of p
l
ai
n
sym
bo
l rec
ogniti
on in
te
xt
for
m
is a g
r
ou
p of i
ndivid
ual
sy
m
bo
ls.
3.3
.
Ap
pli
ca
tio
ns
of N
L
P
In this sect
io
n,
the
ap
plica
ti
ons of
NLP
a
re
di
scusse
d.
3.
3
.1
.
R
ole of
t
he N
LP
in
th
e
fiel
d
of
AI
ap
pli
cat
io
ns
An
arti
fici
al
in
te
ll
igence
(
AI)
is
a
m
achine
w
hich
t
hinks
li
ke
a
hu
m
an
an
d
it
beco
m
es
depend
e
nt
pa
r
t
of
e
ver
y
body’
s
li
fe
by
pro
vid
in
g
a
wi
de
f
orm
of
serv
ic
es
.
Var
i
ous
w
orkis
sti
ll
in
pr
ogress
t
o
m
ake
AI
a
n
eff
ic
ie
nt
syst
em
to
interact
with
hum
an.
On
e
of
the
m
ai
n
te
chn
iq
ues
in
AI
syst
e
m
is
NLP
wh
ic
h
play
s
a
cru
ci
al
ro
le
fo
r
the
com
m
un
ic
at
ion
task
.
W
it
ho
ut
NLP
an
intel
li
gen
ce
syst
em
cann
ot
un
der
sta
nd
the
m
eaning
of
word
s
in
c
on
t
ext
.
So
th
e
NL
P
can
be
cal
le
d
as
fun
dam
ental
un
it
of
the
arti
fici
al
intel
l
i
gen
c
e
wh
ic
h
ena
bles
intel
li
gen
t
syst
e
m
to
interact
eff
ic
ie
ntly
with
hum
ans
by
perform
ing
co
m
m
un
ic
at
ion
proces
s
su
c
h
as
a
utom
at
ed
sp
eec
h
an
d
aut
om
ated
te
xt
wr
it
in
g
b
y
us
i
ng
na
tural
la
ng
uage.
S
om
e
of
the
rece
nt
te
chn
iq
ues
are
discusse
d bel
ow:
a.
NLP
f
or
sp
ee
c
h
sy
nth
esi
s
:
In
this
process
NLP
is
ap
plie
d
for
te
xt
-
to
-
s
peech
c
onversi
on
in
w
hich
te
xt
inf
or
m
at
ion
is
ta
ken
as
in
pu
t
to
the
syst
e
m
a
nd
a
fter
co
nver
sion,
m
achine
rep
l
ie
s
into
s
pe
ech.
T
he
proce
s
s
involves
va
rio
us
high
le
vel
m
od
ules
f
or
pe
rfor
m
ing
s
peec
h
synt
hesis
a
nd
the
sente
nce
unde
rstan
ding
and
segm
entat
ion
pro
ces
s is
done by usi
ng
decisi
on tree.
b.
NLP
f
or
sp
eec
h
rec
ogniti
on
:
In
this
N
LP
in
cl
ud
es
c
on
te
xt
fr
ee
gram
m
ars
to
unde
rstan
d
the
sente
nces
a
nd
phrase
i
n
s
poke
n
la
ng
uag
e
a
nd
the
n
e
xtract
m
eaningfu
l
i
nfor
m
at
ion
an
d
t
ran
s
f
or
m
the
m
to
dig
it
al
patt
er
n
that m
achine can
unde
rstan
d[43
].
3.3
.2
.
R
ole of
t
he N
LP
in
th
e
fiel
d
of
big
-
d
ata and
classi
f
ic
at
io
n
The
te
xt
cl
ass
ific
at
ion
ha
s
c
on
ti
nu
ously
be
en
an
interest
i
ng
t
op
i
c
i
n
th
e
researc
h
fiel
d
of
NLP
,
and
w
hile
ente
rin
g
in
the
a
ge
of
bi
g
data,
ef
f
ic
ie
nt
te
xt
cl
assifi
ers
a
re
diff
i
cult
to
ob
ta
i
n
NLP
f
or
t
he
sc
ie
ntific
data analy
ti
cs. To
provide t
he
h
igh
c
om
pu
ta
ti
on
al
p
e
rfor
m
ance to
te
xt classi
ficat
ion
in
th
e era o
f
bi
g
dat
a [32]
a
deep
ne
ur
al
netw
ork
c
on
ce
pt
is
with
hybr
id
outl
ie
rs
in
wh
ic
h
te
xt
cl
a
ssifie
rs
are
cl
assifi
ed
as
ru
le
s
-
base
d
and
sta
ti
sti
cal
cl
assifi
cat
ion
m
e
tho
ds.
Wh
il
e
m
os
t
of
th
e
m
achine
le
ar
ni
ng
te
ch
niqu
e
s
su
c
h
a
s
s
uppo
rt
vecto
r
m
achine,
decisi
on
tre
e
KNN,
ne
ura
l
network,
ke
r
nel
le
arn
in
g
ha
ve
ap
plied
in
the
process
of
te
xt
cl
assifi
cat
ion
[
44
]
.
3.3
.3
.
NLP in
th
e
are
a of
bu
siness
and
soft
w
are m
anag
e
ment
The
fiel
d
of
business
proc
ess
m
anag
em
ent
faces
a
lot
o
f
c
om
plexiti
es
to
pro
vid
e
qual
it
y
of
ser
vices
du
e
to
the
grow
t
h
of
en
orm
ou
s
dem
and
and
the
te
ch
no
l
og
y.
Ma
ny
researc
her
s
ha
ve
draw
n
at
te
ntion
t
o
red
esi
gn
i
ng
pr
ocess
via
NL
P
to
pro
duce
bette
r
res
ult
an
d
perf
or
m
ances
t
o
the
bu
si
ness
orga
nizat
ion.
A
do
m
ai
n
m
o
del
is
desig
ne
d
to
ha
ndle
the
iss
ue
of
c
us
tom
er
fee
dback
th
r
ough
t
he
s
upport
of
so
ci
al
m
ul
tim
edia
[3
9
]
.
An
NL
P
is
i
m
ple
m
ented
with
geo
l
ocati
on
co
nte
xt
to
achieve
ef
fici
ent
in
te
r
face
by
us
i
ng
fu
zzy
m
od
el
[4
1
]
.
3.
3
.4
.
NLP
f
or
o
n
to
l
ogy l
ear
ning
The
te
rm
“ont
olo
gy”
is
c
onside
red
withi
n
diff
e
ren
t
se
ns
es
i
n
di
ff
e
r
ent
areas
.
F
or
an
a
rtific
ia
l
intel
li
gen
ce,
ontolo
gy
can
be
def
ine
d
as
a
sp
eci
fic
co
ncep
t
su
ch
as
thin
gs,
even
ts
,
and
r
el
at
ion
s
that
are
us
e
d
to
su
pp
or
t
in
form
ation
exc
ha
ng
e
betwee
n
di
ff
e
ren
t
entit
ie
s.
NLP
is
cl
os
el
y
relat
ed
to
the
On
tol
og
y
’s
it
can
be
us
e
d
to
de
fine
essenti
al
r
ules
for
sem
antic
analy
sis
an
d
quest
ion
a
nd
a
nsweri
ng
syst
em
.
Va
rio
us
a
ppr
oa
ches
su
c
h
as
sy
m
bo
li
c,
sta
ti
sti
cal
and
hy
br
i
d
hav
e
dr
a
w
n
to
inf
or
m
at
i
on
e
xtracti
on,
and
r
et
rieval
fr
om
the
knowle
dge
resour
ce
a
nd
s
om
e
of
these
te
chn
i
qu
e
s
ha
ve
been
ta
ken
i
nto
c
on
si
der
at
i
on
for
te
xt
m
in
ing
via
on
t
ology
le
ar
ni
ng
a
nd
w
hich
is
ha
ving
an
eff
ect
ive
res
ult.
The
sym
bo
li
c
m
et
ho
d
e
xp
loit
s
hu
m
an
na
tural
la
nguag
e
(lin
guist
ic
inf
or
m
ation)
da
ta
to
m
ine
use
f
ul
in
form
ation
f
r
om
t
he
te
xt.
S
uch
as
noun
phras
es
are
ta
ken
a
s to
lexi
cal
idea to
d
e
note
in
a
n o
nto
l
og
y.
3.2.5
.
NLP in
th
e
are
a of
Bi
oinform
at
ic
s
Bi
oin
f
or
m
at
ic
s
is
cl
ass
of
sci
ence
w
hich
use
s
the
com
pu
te
r
te
chnol
og
y
to
extra
ct
the
knowl
ed
ge
or
inf
or
m
at
ion
from
the
colle
ct
ion
of
bio
l
og
ic
al
data.
The
proc
ess
of
in
form
at
io
n
extracti
on
i
nvolv
e
s
the co
ll
ect
ion,
stora
ge,
m
anip
ulati
on
,
m
od
el
ing an
d retrie
va
l for a
naly
zi
ng
,
p
re
dicti
ng and
v
isuali
zi
ng th
r
ough
the
de
velo
pm
e
nt
of
s
of
t
war
e
an
d
the
al
gorithm
.
Text
m
ining
[
41
]
a
nd
NLP
te
c
hn
i
qu
e
s
play
the
very
i
m
po
rtant
r
ole
to r
ec
over
us
e
r
prefe
ren
ce
kn
ow
le
dg
e
fro
m
g
r
ow
i
ng
datab
ases.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol. 10,
No.
5,
Oct
ob
e
r 2
020
:
49
28
-
49
36
4934
3.2.6
.
Te
xt mi
ning ap
proac
hes to
protei
n
research
DNA
an
d
prot
ei
n
sequ
e
nces
are the
ty
pes
of g
eneti
cal
lan
gu
a
ge
co
des
in
w
hich
N
LP and text
m
ining
m
et
ho
ds
are
a
pp
li
ed
to
analy
ze
the
bio
i
nform
at
ic
s.
A
la
te
nt
sem
antic
analy
sis
was
a
ppli
ed
to
protei
n
rem
ote
ho
m
olo
gy
de
te
ct
ion
,
a
nd
a
na
ly
sis
of
prot
ei
n
sp
ect
ra
l
ori
gin
at
es
f
r
om
the
sta
ti
sti
ca
l
fr
eq
ue
ncy
in
NLP
,
and
var
io
us
w
eb
se
r
ver
s
we
r
e
desi
gn
e
d
to
extract
knowle
dg
e
feat
ur
es
a
nd
gram
m
ar
rul
es
of
protei
n,
DNA,
and
RN
A
se
qu
ences
.
I
n
t
he
fiel
d
of
predic
ti
ng
protei
n
st
r
uctu
re,
s
om
e
research
w
orks
ha
ve
us
ed
m
a
ch
ine
le
arn
in
g
pr
e
di
ct
ion
m
et
ho
ds
for
pr
e
dicat
io
n
of
protei
n
di
so
r
der
w
hich
base
d
on
AI
a
nd
ne
ur
al
net
work,
m
axi
m
u
m
entro
py,
s
upport
vecto
r
m
achine
and
ra
ndom
fo
rest
.
F
or
predict
io
n
of
protei
n
f
unct
ion
so
m
e
ano
t
her
w
ork
is
carrie
d
out
in
te
rm
s
of
ide
ntifyi
ng
reg
i
on
of
s
ub
-
cel
lul
ar
protei
n,
pr
otein
rem
ote
hom
olo
gy
detect
iono
bs
er
ving
m
ulti
fu
nc
ti
on
al
enzym
e
s,
cl
assify
in
g
protei
n
functi
ons
a
nd
cl
assif
yi
ng
tra
ns
-
m
em
br
ane
protei
n
by
us
i
ng
m
achine
le
arn
i
ng
al
gorith
m
su
ch
as
SVM
,
kernel
m
e
thod,
decisi
on
-
m
aking
tree
,
r
andom
forest,
Ba
ye
sia
n
netw
ork
Tex
t
pr
ocessi
ng
m
et
hods
w
hich
a
re
as
on
tol
og
y
annotat
ion
a
nd
sa
m
ple
weight
ing
,
are
use
d
to
de
te
ct
featur
es
and
proce
ss
tr
ai
nin
g
data
a
nd
gr
am
m
atical
analy
ses
hidden
Ma
r
kov
m
od
el
(H
MM
)
[
38, 3
9]
.
4.
KEY
CHALL
ENGES I
N N
LP
The
key c
halle
ng
e
s in
N
L
P a
r
e:
a.
High
Scal
e
I
nfor
m
at
ion
Ac
qu
isi
ti
on
:
For
the
pr
e
dicti
on
of
l
inguist
ic
struct
ur
e
i
nfo
rm
ation
,
knowle
dge
of
idiom
s,
le
xical
and
ge
ner
al
pa
tt
ern
s
rec
ogni
ti
on
,
m
os
tl
y
th
ou
s
an
ds
of
m
illi
on
s
of
knowle
dg
e
i
nfor
m
at
io
n
will
b
e
require
d.
b.
Am
big
uity
Pr
ob
le
m
:
A
m
bi
gu
it
y
is
the
m
os
t
sign
ific
a
nt
chall
eng
e
i
n
NLP.
NLs
are
ridd
le
d
wit
h
a
m
big
uiti
es
at
each
le
vel
of
descr
i
ption,
i.e
.,
f
ro
m
phonet
ic
to
so
ci
ologic
al
.
Un
ti
l
no
w,
hu
m
an’
s
natu
r
al
la
nguag
e
s,
syst
e
m
is
un
awa
re
of
t
his
pe
rv
asi
ve
am
big
uity
—
i
t
com
es
to
research
e
r’s
at
te
ntion
i
n
the
guise
of su
c
h
m
isund
erstan
dings,
lo
gisti
cal
m
arg
in
al
co
nce
pt a
nd
con
te
ste
d
li
bel
su
it
s.
c.
Lan
gu
a
ge
Va
r
ia
bili
ty
:
Hu
m
a
n
la
ngua
ges
a
re
highly
rich
,
and
t
ho
se
ha
ve
dif
fer
e
nt
ways
to
ex
pr
e
ss
a sp
eci
fic
m
ea
ning li
ke,
e.g.
“wh
ic
h
is t
he
near
est
h
otel
?
” & “can
you
ple
ase te
ll
m
e
the
address of
nea
r
est
ho
te
l
?
”
bo
t
h
se
ntences
ha
ve
a
si
m
il
ar
m
eani
ng.
T
hus,
NLP
ap
proach
will
fig
ur
e
out
th
ose
two
sente
nc
es
m
ean
the
sam
e.
T
her
e
fore,
t
her
e
are
seve
r
al
ways
to
re
present
t
he
te
xt
sem
a
ntics
or
to
disti
ng
uish
two
te
xts.
d.
Sem
antic
s
Pr
oble
m
:
The
f
un
dam
ental
chall
eng
e
is
that
we
don’t
hav
e
the
knowle
dge
of
hu
m
ans
sem
antic
structu
res
a
re
li
ke;
on
e
si
gn
i
ficant
m
et
ho
d
is
to
stud
y
t
hat
they
are
ve
ry
near
e
r
to
hu
m
an
synta
ct
ic
structu
res.
I
n
t
his
m
e
tho
d,
a
su
it
able
stud
y
of
sem
antic
s
is
s
uppose
d
to
pro
vid
e
hi
gh
c
onstrai
nt
the
le
vel of
interp
retat
ion
s
.
e.
Kno
wled
ge
Re
pr
ese
ntati
on
:
How
do
hum
a
n
re
present
t
he
bac
kgrou
nd
w
or
l
d
in
form
at
io
n,
by
pointi
ng
tha
t
it
req
uires
to
i
nteract
with
th
e
structu
re
of
s
e
m
antic
rep
res
entat
ion
in
the
la
nguag
e
unde
rstan
ding
proc
ess.
Hav
e
to
a
naly
ze
that
kn
owl
edg
e
a
nd
m
ea
ning
represe
nt
at
ion
is
the
sam
e,
it
m
eans
that
la
ng
ua
ge
is
a m
irror
of
m
ind
or m
ental
r
ep
resen
ta
ti
on
of the
world
.
f.
Learn
i
ng
of
la
ngua
ge:
Wh
at
are
the
words,
and
how
t
hey
are
cat
eg
or
iz
e
d
and
w
hat
is
it
s
patte
rn
struct
ur
e
(i.e.,
i
dio
m
s,
synta
ct
ic
,
et
c.)
and
is
a
ver
y
com
plex
ta
sk
,
therefo
re
the
ro
le
of
la
ngua
ge
le
ar
ning
a
nd
knowle
dge
representat
io
n
is
one
of the c
omm
on
ch
al
le
nge
s in NLP
app
roach.
5.
CONCL
US
I
O
N
Fr
om
this
stu
dy
can
c
on
cl
ude
that
NL
P
is
a
sign
i
ficant
a
ppr
oach
w
hich
has
uniq
ue
f
e
at
ur
es
with
excell
ent
c
omm
un
ic
at
ion
ap
proac
h,
i.e
.,
N
LP
is
t
he
set
of
to
ols
a
nd
m
et
ho
dolo
gy
of
kn
ow
i
ng
to
achie
ve
the
goal
s
a
nd
get
res
ults
dur
ing
knowle
dge
disc
ov
e
ry.
T
he
r
ef
or
e
,
the
ca
se
stu
dy
pr
ov
i
des
t
he
necess
it
y
of
the
NL
P
in
the
c
urren
t
c
om
pu
ti
ng
w
orl
d
al
on
g
with
diff
e
re
nt
ap
proach
e
s
a
nd
t
he
ir
ap
plica
ti
ons.
Als
o,
it
h
igh
li
ghts t
he
criti
cal
ch
al
le
ng
e
s in
the
de
ve
lop
m
ent o
f
n
e
w NLP m
od
el
.
REFERE
NCE
S
[1]
H
.
La
n
e,
e
t al
.
,
“
Natur
al Languag
e
Proce
ss
ing
in A
ct
ion,
”
Mann
in
g
Publications
,
2
018
.
[2]
S
.
Singh,
“
The
role
of
spee
ch
te
chno
log
y
in
biometri
cs,
fore
nsics
and
m
an
-
m
ac
hine
int
erf
a
ce
,
”
Int
ernati
on
al
Journal
of
Elec
t
rical
and
Computer
Eng
ine
ering
(
IJE
CE)
,
v
ol. 9,
n
o.
1
,
pp
.
281
-
2
88,
2019
.
[3]
I
.
El
Baz
i
and
N
.
La
a
chf
oubi
,
“
Arabi
c
Nam
ed
Ent
ity
Re
cogni
t
i
on
Us
ing
Dee
p
Le
arn
ing
Approac
h
,
”
Inte
rnatio
nal
Journal
of
Elec
t
rical
and
Computer
Eng
ine
ering
(
IJE
CE)
,
v
ol. 9,
n
o.
3,
pp
.
2025
-
2032,
2019
.
[4]
Prathe
ek
I
.
and
J.
Paulose
,
“
Predic
ti
on
o
f
Ans
wer
Ke
y
words
using
Char
-
RNN
,
”
Inte
rnational
Journ
al
of
El
ectric
a
l
and
Computer
E
ngine
ering
(
IJECE)
,
v
ol. 9,
n
o.
3,
pp
.
2164
-
217
6,
2019
.
[5]
“
Natur
al
La
ngua
ge
Proce
ss
ing
,
”
[
Online
]
.
Avai
la
b
le
:
htt
ps:
//
en
.
wik
ipe
di
a.
org/wik
i/N
at
ura
l_
la
nguag
e_pr
oce
ss
ing.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N:
20
88
-
8708
A p
ri
or
c
as
e
stud
y
o
f
natur
al
langua
ge pr
oc
essing o
n dif
fe
rent
dom
ain
(
Shru
t
hi J
.
)
4935
[6]
“
Sy
nt
ax
-
Engl
is
h
sente
n
ce struc
t
ure
,
”
[Onlin
e]
.
Avail
ab
le
:
esl
.
fis
.
edu/lea
rn
ers/a
d
vic
e
/s
y
nt
ax.
h
tm.
[7]
Rat
inov
,
L
ev,
an
d
Dan
Roth
,
“
De
sign
challe
ng
es
and
m
isconc
ep
ti
ons
in
named
en
ti
t
y
rec
ogn
it
ion
,
”
In
Proceedi
ng
s
of
th
e
Thir
teenth
Confe
ren
ce on Com
putat
ional
Natural
Language
Learning
(
CoNL
L
-
2009
)
,
pp.
147
-
1
55,
2009
.
[8]
R
.
J.
Moone
y
and
R
.
Bun
esc
u,
“
Mining
Know
le
dge
from
Te
xt
Us
ing
Inf
orm
at
ion
Ext
r
a
ct
ion
,
”
SIGKD
D
Ex
plorati
ons
,
v
o
l.
7
,
no.
1
,
pp
.
3
-
10
,
2005
.
[9]
M
.
Coll
ins,
“
C
hapt
er
2
:
Ta
gg
i
ng
Problems
,
and
Hidden
Markov
Models,
”
Course
note
for
NLP,
Colum
bia
Univer
sit
y
,
2013
.
[10]
M
.
La
p
at
a
and
F
.
Keller,
“
W
eb
-
base
d
m
odel
s
for
Natur
al
La
ngu
ag
e
Proce
ss
ing
,
”
A
CM
Tr
ansacti
ons
on
spee
ch
and
Language
Proces
sing
,
v
ol
.
2
,
n
o.
1
,
pp
.
1
-
30,
Feb
.
2005
.
[11]
N
.
Ranjan,
et
al.,
“
A
Surve
y
on
Te
chn
ique
s
in
NLP
,
”
Int
ernational
Journal
of
Computer
Appl
i
cat
ions
,
v
o
l.
13
4
,
n
o.
8
,
pp
.
6
-
9
,
Ja
n
.
2016
.
[12]
A
.
Copesta
k
e
,
“
Natur
al Languag
e
Proce
ss
ing
,
”
[
Online
]
.
Avai
la
b
le
:
h
tt
p
:/
/www
.
c
l
.
ca
m
.
ac.uk/
users
/a
a
c/
.
[13]
R.
V
.
Alm
ada
,
e
t
al
.
,
“
Natur
al
L
angua
ge
Proc
essing
and
Te
xt
Mi
ning
to
Ide
nti
f
y
Know
le
dge
Profile
s
for
Software
Engi
ne
eri
ng
Pos
it
ions:
Gen
era
t
i
ng
Know
le
dge
Profile
s
from
Resum
es,
”
2017
5th
Inte
rnat
iona
l
Confe
ren
ce
in
Soft
ware
Engi
ne
ering
R
ese
arch
and
Innov
at
ion
(
CONISO
FT
)
,
Mérida,
Mex
ic
o
,
p
p.
97
-
106
,
2017
.
[14]
C.
B.
Q.
Pelay
o
,
et
al
.
,
“
Natur
al
la
nguag
e
proc
es
sing
for
improving
hea
ring
i
m
pa
ire
d
student
re
ad
ing
skill
s
,
”
2017
Inte
rnational
Co
nfe
renc
e
on
Info
rm
ati
on
Syste
ms
and
Computer
S
ci
en
ce (
INCISCOS)
,
Quito
,
pp
.
201
-
206
,
2017
.
[15]
K.
Khali
d
,
e
t
a
l.
,
“
Ext
ension
of
Sem
ant
ic
Bas
ed
Urdu
Li
nguisti
c
Resourc
es
Us
i
ng
Natur
al
L
angu
age
Proc
essing,
”
2017
IEE
E
15
th
Int
ernati
onal
C
onf
ere
nc
e
on
De
pendabl
e
,
Aut
on
omic
and
Sec
ur
e
Computing,
1
5th
Int
ernati
ona
l
Conf
ere
nc
e
on
Pe
rvasiv
e
Inte
l
ligence
and
Computing,
3rd
Int
e
r
nati
onal
Conf
er
enc
e
on
Bi
g
Dat
a
Inte
ll
ig
ence
and
Computing
and
Cybe
r
Sci
en
ce
and
Technol
ogy
Congress(
DAS
C/P
iCom/DataC
om/Cybe
rSciTe
c
h)
,
O
rlando,
FL
,
pp.
1322
-
1325
,
2017
.
[16]
J.
L.
J
.
Marque
z,
et
al
.
,
“
Chall
enge
s
And
Opportunit
i
es
In
An
aly
tic
-
Predi
ct
iv
e
Envi
ronm
ent
s
Of
Big
Data
And
Natur
al
L
angua
g
e
Proce
ss
ing
For
Socia
l
Network
Rat
ing
S
y
st
ems
,
”
in
IE
EE
Latin
Ame
rica
Tr
ans
act
ions
,
vol
.
16,
no.
2
,
pp
.
592
-
5
97,
Feb
.
2018
.
[17]
H.
El
ah
ini
a
,
e
t
al
.
,
“
Predic
t
in
g
m
edi
ca
l
non
a
dher
ence
using
nat
ura
l
la
ngu
a
ge
proc
essing
,”
2017
IEE
E
M
IT
Undergr
aduate
Re
search
Te
chn
ology
Con
fe
renc
e
(
URTC)
,
Camb
ridge
,
MA
,
pp
.
1
-
4
,
2017
.
[18]
S.
Slat
er
,
et
al.
,
“
Us
ing
nat
ura
l
language
proc
essing
tool
s
to
devel
op
comple
x
m
odel
s
of
student
en
gage
m
ent
,
”
201
7
Sev
en
th
Inte
rnat
ional
Confe
renc
e
on
Af
fe
c
tive
Computing
and
Inte
lligen
t
Inte
r
act
ion
(
ACII)
,
S
an
Antonio,
TX
,
pp.
542
-
547
,
20
17
.
[19]
I.
C
alapode
scu
,
et
a
l.
,
“
Sem
i
-
Autom
at
ic
De
-
id
enti
fic
a
ti
on
of
Hos
pi
ta
l
Disch
arg
e
Su
m
m
ari
es
with
Natur
a
l
La
ngu
age
Proce
ss
ing:
A
Case
-
Stud
y
of
Perfor
m
anc
e
an
d
Rea
l
-
W
orld
U
sabil
ity
,
”
2017
I
EE
E
Int
ernati
on
al
Confe
renc
e
on
Inte
rnet
of
Thin
gs
(
i
Things
)
an
d
IEE
E
Gr
ee
n
Computing
and
Comm
unic
ati
o
ns
(Gr
ee
nCom
)
and
IEE
E
Cyb
er,
Phy
sica
l
and
So
ci
al
Computing
(
CPSCom)
and
IEE
E
Smar
t
Data
(
Smar
tData
)
,
E
xet
er
,
pp
.
1106
-
1111
,
2017
.
[20]
P.
Bronie
cki
an
d
A.
Hanc
har
,
“
Data
Innova
t
ion
for
Inte
rna
ti
on
a
l
Deve
lopment:
An
over
vie
w
of
nat
ura
l
la
nguag
e
proc
essing
for
q
ual
itati
v
e
da
ta
a
naly
s
is
,”
2
017
I
nte
rnational
Confe
renc
e
on
th
e
Fronti
ers
and
A
dvanc
es
in
Da
ta
Sci
en
ce (
FA
DS)
,
Xi'an
,
pp.
9
2
-
9
7
,
2017
.
[21]
S.
Ži
tni
k
,
e
t
al.
,
“
nutIE
—
A
m
od
ern
open
source
nat
ur
a
l
la
ngu
age
proc
essing
toolkit
,”
2017
25th
Telecomm
unic
atio
n
Forum (
TEL
FO
R)
,
pp.
1
-
4
,
2017
.
[22]
C.
J.
Bab
y
,
e
t
al.
,
“
Hom
e
aut
om
at
ion
using
IoT
a
nd
a
chatbot
usi
ng
nat
ura
l
l
a
ngu
age
proc
essing
,”
2017
Innov
ati
o
n
s
in
Powe
r and
Ad
vanc
ed
Computi
ng
Technol
og
ie
s
(
i
-
PA
CT
)
,
pp.
1
-
6,
2017
.
[23]
A.
R.
Sharm
a
and
P
.
Kaushik
,
“
Li
te
r
at
ure
surve
y
of
stat
isti
ca
l
,
de
e
p
and
rei
nfor
cem
ent
le
arn
ing
in
nat
ura
l
la
nguag
e
proc
essing
,”
20
17
Inte
rnation
al
Confe
ren
ce
on
Computin
g,
Comm
unic
at
io
n
and
Aut
o
mation
(
ICCCA)
,
pp.
350
-
354
,
20
17.
[24]
M.
A.
Ahm
ed
and
S
.
Tra
usan
-
Ma
tu
,“
Us
ing
nat
ura
l
la
nguag
e
proc
e
ss
ing
for
ana
l
y
z
i
ng
Arabi
c
poe
tr
y
rh
y
thm
,”
2017
16th
Ro
EduNe
t
Confe
renc
e
N
etwor
ki
ng
in
Education
and
Re
s
ea
rch
(
RoE
duNet
)
,
pp.
1
-
5
,
2017.
[25]
K.
R.
Pole
and
V
.
R.
Mote
,
“
I
m
provised
fuz
z
y
c
luste
ring
usi
ng
name
ent
i
t
y
rec
ogni
ti
on
and
nat
ura
l
la
ngu
ag
e
proc
essing
,”
20
17
1st
Inte
rnat
i
onal
Confe
ren
ce
on
Intelli
g
ent
Syste
ms
and
Inf
orm
ati
on
Manage
ment
(
ICI
SIM)
,
pp.
123
-
126
,
20
17.
[26]
C.
Ber
te
ro
,
e
t
al
.
,
“
Expe
r
ie
n
ce
Report
:
Log
Mining
Us
ing
Natur
al
La
ngu
a
ge
Proce
ss
ing
a
nd
Ap
pli
c
at
ion
to
Anom
aly
De
te
c
ti
on,
”
2017
IE
E
E
28th
Inte
rnat
ional
Symposiu
m
on
Soft
ware
Reliabi
l
ity
En
gine
ering
(
ISSRE
)
,
Toul
ouse,
pp.
35
1
-
360
,
2017
.
[27]
M.
S.
M.
Suhai
m
in,
et
a
l.
,
“
Nat
ura
l
la
nguag
e
pr
oce
ss
ing
base
d
fea
tur
es
for
sar
c
asm
det
ection:
An
inve
stig
at
ion
using
bil
ingu
al
s
oci
a
l
m
edi
a
t
exts
,
”
2017
8th
Inter
nati
onal
Conf
ere
nce
on
Informa
ti
on
Techno
logy
(
ICIT
)
,
Amm
an,
pp.
703
-
709
,
20
17
.
[28]
N.
Alami,
e
t
a
l.
,
“
A
sem
i
-
aut
oma
te
d
appr
oa
ch
f
or
gene
r
at
ing
se
quenc
e
d
ia
gr
ams
from
Arabi
c
user
req
uir
ements
using
a
na
tura
l
l
angua
ge
proc
essing
tool
,
”
2017
8
th
Inte
rnational
Confe
renc
e
on
Information
Technol
ogy
(
ICIT
)
,
Amm
an,
pp.
309
-
314
,
2017
.
[29]
H.
Zhua
ng,
et
al
.
,
“
Natur
al
Language
Proce
ss
i
ng
Servic
e
Bas
ed
on
Stroke
-
L
eve
l
Convolu
ti
o
nal
Networks
for
Chine
se
Te
x
t
Cla
ss
ifi
c
at
ion
,
”
2
017
IEE
E
Int
ernati
onal
Conf
ere
nce
on
We
b
Serv
ices
(
ICWS
)
,
Honolulu,
HI
,
pp.
404
-
411
,
20
17
.
[30]
S.
Deshm
ukh,
et
al
.
,
“
Sia:
An
in
te
ra
ct
iv
e
m
edi
c
a
l
assistant
usi
ng
nat
ura
l
la
ngu
ag
e
proc
essing,
”
20
16
Inte
rnational
Confe
renc
e
on
Global
Tr
ends
in
Signal
Proc
essing,
Informatio
n
Computing
and
Comm
unic
ati
o
n
(
ICG
TSPICC),
Jal
gaon,
pp.
584
-
586
,
2016
.
[31]
T.
Eft
imov,
e
t
a
l.
,
“
A
rule
-
base
d
named
-
ent
ity
rec
ogni
ti
on
m
et
hod
for
knowledge
ext
ra
ct
ion
o
f
evi
denc
e
-
base
d
die
t
ar
y
r
ec
om
m
enda
ti
ons
,”
Pl
oS
one
,
vo
l.
12
,
no.
6
,
p
p.
e0179488
,
2017
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
20
88
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol. 10,
No.
5,
Oct
ob
e
r 2
020
:
49
28
-
49
36
4936
[32]
J
.
Jiang,
“
Infor
m
at
ion
Ext
r
ac
t
io
n
from
Te
xt
,
”
i
n
Aggarwal
C.
and
Zhai
C.
(
ed
s)
,
“
Mining
Te
x
t
Data
,”
Spring
e
r
,
Boston,
MA
,
pp
.
11
-
41,
2012
.
[33]
S.
Boula
knade
l
,
et
al.
,
“
Am
az
ighe
Nam
ed
Ent
ity
Rec
ognition
using a
A rul
e
base
d
appr
oac
h
,
”
2014
IEE
E/
ACS
11t
h
Inte
rnational
Co
nfe
renc
e
on
Co
mputer
Syste
ms
and
Applications (
AICCSA
)
,
Doha,
pp
.
478
-
4
84
,
2
014
.
[34]
H.
El
sa
y
e
d
and
T
.
El
gha
zal
y
,
“
A
named
en
ti
t
ies
rec
ogni
ti
on
s
ystem
for
m
oder
n
standa
rd
Arabi
c
using
rul
e
-
base
d
appr
oac
h
,”
2015
Fi
rs
t
In
te
rnatio
nal
Conf
ere
nce
on
Arabic Computat
ional Lingui
stic
s (
ACLing)
,
pp.
51
-
54
,
2015
.
[35]
Za
y
kovski
y
,
Dm
it
r
y
,
“
Surve
y
of
the
spee
ch
re
cog
nit
ion
techniqu
e
s
for
m
obil
e
devi
ce
s,
”
Proc.
of
DS
Publ
ic
ati
ons
,
2
006
.
[36]
R.
S.
Chav
an
a
nd
G
.
S.
Sabl
e
,
“
An
Overvi
ew
o
f
Speec
h
Recogniti
on
Us
ing
HM
M
,”
Inte
rna
ti
onal
Journal
o
f
Computer
Scien
ce
and
Mobi
le Computi
ng
,
vo
l.
2
,
no.
6
,
pp.
233
-
2
38
,
2013
.
[37]
T.
B
.
Am
in
and
I
.
Mahm
ood
,
“
Speec
h
rec
ogn
it
io
n
using
d
y
n
amic
ti
m
e
warpi
ng
,”
2nd
Inte
rnationa
l
Confe
ren
ce
o
n
Adv
anc
es
in
Spa
ce
Te
chnol
ogi
es,
2008
(
ICAST 20
08
),
pp.
74
-
7
9
,
2
008.
[38]
M.
Sale
em,
et
a
l.
,
“
Self
learni
n
g
spee
ch
r
ec
ogn
it
ion
m
odel
usin
g
vec
tor
qu
antiz
at
ion
,
”
2016
S
ixth
Inte
rnat
ional
Confe
renc
e
on
I
nnovat
i
ve
Comp
uti
ng
Te
chnol
og
y
(
INTECH)
,
Du
bli
n
,
pp.
199
-
20
3
,
2016
.
[39]
S.
T.
Pan,
et
al.
,
“
Speec
h
rec
ogni
ti
on
via
Hidden
Markov
Model
and
neur
al
net
wo
r
k
tra
ine
d
b
y
ge
net
i
c
al
gori
thm,
”
2010
Inte
rnat
ion
al
Conf
ere
nce o
n
Ma
chi
n
e
Lear
ning
and
C
ybe
rn
et
i
cs,
Qingd
ao,
p
p.
2950
-
2955
,
2
010
.
[40]
P.
P.
Pata
nge
a
nd
J.
S.
R.
Alex,
“
Im
ple
m
ent
ation
of
AN
N
base
d
spee
ch
rec
o
gnit
i
on
s
y
st
em
on
an
embedde
d
boar
d,
”
2017
Int
ernati
onal
Conf
ere
nce
on
Next
gen
El
e
ct
ronic
Technol
ogi
es:
Sil
ic
on
to
Sof
twar
e
(
ICNETS2
)
,
Chenna
i
,
pp
.
40
8
-
412
,
2017
.
[41]
A
.
Chaudhuri,
e
t
al.
,
“
Optical
Chara
c
te
r
R
ec
o
gnit
ion
S
y
stems
for
Diffe
ren
t
L
angua
ges
wi
th
Soft
Com
puti
ng,
”
Springer
,
2016
.
[42]
A.
Reshamw
al
a
,
et
al.,
“
Revi
ew
on
na
tural
l
ang
uage
pro
ce
ss
ing
,”
IR
ACST
Engi
n
ee
ring
S
ci
en
ce
a
nd
Technol
og
y:
An
Int
ernati
onal
Journal
(
ESTIJ)
,
vol
.
3
,
no
.
1
,
pp
.
113
-
116
, 2
013
.
[43]
A.
Trilla,
“
Nat
ura
l
La
nguag
e
Proce
ss
ing
t
e
chni
ques
in
T
e
xt
-
To
-
Spee
ch
s
y
nth
esis
and
Autom
at
ic
Speech
Rec
ognition
,”
Depa
rt
ament
de
T
ec
nolog
ie
s
Media
Engi
n
y
eria
i
Arquite
c
tura
La
Sall
e
(Unive
rsit
at
Ramon
Ll
ull
)
,
Barc
e
lona
,
2009
.
[44]
C.
Strappa
r
ava
and
R.
Mihalce
a
,
“
Le
arn
ing
to
i
dent
if
y
emot
ionsi
nte
xt
,
”
in
S
AC’0
8:
Proce
edi
ngs
of
the
2008
ACM
sympos
ium
on
A
ppli
ed
computi
n
g,
New York
,
pp
.
1556
-
1560
,
20
08.
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Sh
ruthi
J
.
completed
B.
E
i
n
CS
E
from
VTUin
t
he
y
e
ar2
006,
a
n
d
complet
e
d
Mte
ch
from
VTU
in
the
y
ear
2008
.
Curre
n
tly
purs
uing
Ph.D
in
C
om
pute
r
Scie
n
ce
and
Eng
ineeri
n
g,
Sir
MV
IT
under
VTU.Cur
ren
tly
working
as
As
sistant
Profess
or,
Depa
rtment
of
CS
E,
BMS
ITM,
Benga
luru
,
with
10
y
e
ars
of
te
a
chi
ng
expe
r
ie
nc
e
and
al
so
has
0
1
y
ea
r
of
industr
y
expe
r
ie
n
ce.
Area
s of
intere
st
are
Spe
ec
h
Proc
essing,
Dat
a
m
ini
ng,
Big
Data,
M
ac
hin
e
L
ea
rning
and
Natur
al
La
nguag
e
Proc
e
ss
ing.
Dr.
Su
ma
Sw
a
my
complet
ed
h
er
B.
E
in
ECE
fr
om
KBP
Coll
ege
of
Engi
n
ee
ring
,
Sata
r
a
under
Shivaj
i
Univ
ersi
t
y
,
and
Kolh
apur
in
1990
.
She
co
m
ple
te
d
Pos
t
Gr
adua
t
e
Diploma
in
Advanc
ed
Com
pute
r
Te
chn
olog
y
(AS
SET)
from
Aptec
h,
M
G
Road,
Bengaluru
in
1996.
Sh
e
completed
her
M.T
ec
h
in
E
CE
from
Sir
MV
IT,
Beng
al
uru
under
VTU
in
2
005.
She
comple
te
d
her
Ph.D
in
Inform
at
ion
and
Com
m
unic
at
ion
Eng
ine
er
in
g,
Anna
Univer
sit
y
Ch
ennai
in
2014.
She
is
cur
ren
t
l
y
workin
g
as
Profess
or,
Depa
rtment
of
CS
E,
Sir
MV
IT,
Benga
luru.
She
is
guidi
ng
6
rese
ar
ch
schol
ar
s
under
VTU.
S
he
has
27+
y
e
ar
s
of
expe
rie
nc
e
in
te
a
chi
ng
with
2
y
ea
rs
of
industr
y
expe
r
ience
.
Her
ar
ea
s
of
int
er
est
ar
e
Speec
h
Proc
essing,
Dat
a
m
ini
n
g,
Big
Da
ta,
Mac
hine
Learni
ng,
Natur
al
La
n
guage
Proc
essing,
Algori
thms
,
Data
base
Mana
g
ement
S
y
st
e
m
s
and
IoT
.
She
h
a
s
21
publi
c
at
ion
s
in
her
cre
d
it
in
ren
owned
p
ee
r
rev
ie
wed
Journa
ls
with
g
ood
impact
f
ac
tor
a
nd
Nati
on
al
an
d
Inte
rn
at
ion
al
Confer
ences.
Sh
e
has
al
so
au
th
ore
d
a
book
“
PR
ACTICAL
AP
PLICATIONS
OF
SP
EE
CH
SIG
NA
LS
"
pu
bli
shed
b
y
LAP
LAMBERT,
Germ
a
n
y
.
She
h
as
42
ci
tations
f
or
her
publicatio
ns
with
h
inde
x
of
2
with
top
h
c
it
ed
r
ese
arc
h
publi
c
at
ion
ti
t
led
“
Speake
r
Inde
pende
n
t
Digit
Rec
ognition
S
y
s
te
m
“
(18
ci
ta
t
io
ns)
and
RG
score
of
1.
29.
She
is r
evi
ewe
r
for m
an
y
IEEE
Inte
r
nat
ion
al
Journals
.
She
was S
essio
n
Chai
r
for
m
an
y
Inte
rn
at
io
nal
and
Nati
ona
l
Confer
ences.
She
is
an
Edi
torial
Board
m
emb
er
of
Scie
nce
Publishing
Group,
US
A.
She
is a L
if
e
Mem
ber
of
CS
I
and
ISTE
p
rofe
ss
iona
l
bod
i
es.
Evaluation Warning : The document was created with Spire.PDF for Python.