<s>
In	O
natural	B-Language
language	I-Language
processing	I-Language
,	O
entity	B-General_Concept
linking	I-General_Concept
,	O
also	O
referred	O
to	O
as	O
named-entity	O
linking	O
(	O
NEL	O
)	O
,	O
named-entity	O
disambiguation	B-General_Concept
(	O
NED	O
)	O
,	O
named-entity	B-General_Concept
recognition	I-General_Concept
and	O
disambiguation	B-General_Concept
(	O
NERD	O
)	O
or	O
named-entity	O
normalization	O
(	O
NEN	O
)	O
is	O
the	O
task	O
of	O
assigning	O
a	O
unique	O
identity	O
to	O
entities	O
(	O
such	O
as	O
famous	O
individuals	O
,	O
locations	O
,	O
or	O
companies	O
)	O
mentioned	O
in	O
text	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
is	O
different	O
from	O
named-entity	B-General_Concept
recognition	I-General_Concept
(	O
NER	O
)	O
in	O
that	O
NER	O
identifies	O
the	O
occurrence	O
of	O
a	O
named	B-General_Concept
entity	I-General_Concept
in	O
text	O
but	O
it	O
does	O
not	O
identify	O
which	O
specific	O
entity	O
it	O
is	O
(	O
see	O
Differences	O
from	O
other	O
techniques	O
)	O
.	O
</s>
<s>
In	O
entity	B-General_Concept
linking	I-General_Concept
,	O
words	O
of	O
interest	O
(	O
names	O
of	O
persons	O
,	O
locations	O
and	O
companies	O
)	O
are	O
mapped	O
from	O
an	O
input	O
text	O
to	O
corresponding	O
unique	O
entities	O
in	O
a	O
target	O
knowledge	O
base	O
.	O
</s>
<s>
Words	O
of	O
interest	O
are	O
called	O
named	B-General_Concept
entities	I-General_Concept
(	O
NEs	O
)	O
,	O
mentions	O
,	O
or	O
surface	O
forms	O
.	O
</s>
<s>
The	O
target	O
knowledge	O
base	O
depends	O
on	O
the	O
intended	O
application	O
,	O
but	O
for	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
intended	O
to	O
work	O
on	O
open-domain	O
text	O
it	O
is	O
common	O
to	O
use	O
knowledge-bases	O
derived	O
from	O
Wikipedia	O
(	O
such	O
as	O
Wikidata	O
or	O
DBpedia	B-Application
)	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
techniques	O
that	O
map	O
named	B-General_Concept
entities	I-General_Concept
to	O
Wikipedia	O
entities	O
are	O
also	O
called	O
wikification	O
.	O
</s>
<s>
Considering	O
again	O
the	O
example	O
sentence	O
"	O
Paris	O
is	O
the	O
capital	O
of	O
France	O
"	O
,	O
the	O
expected	O
output	O
of	O
an	O
entity	B-General_Concept
linking	I-General_Concept
system	O
will	O
be	O
Paris	O
and	O
France	O
.	O
</s>
<s>
(	O
URIs	B-Protocol
)	O
for	O
the	O
entities	O
in	O
the	O
knowledge	O
base	O
.	O
</s>
<s>
Using	O
a	O
different	O
knowledge	O
base	O
will	O
return	O
different	O
URIs	B-Protocol
,	O
but	O
for	O
knowledge	O
bases	O
built	O
starting	O
from	O
Wikipedia	O
there	O
exist	O
one-to-one	O
URI	B-Protocol
mappings	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
is	O
a	O
critical	O
step	O
to	O
bridge	O
web	O
data	O
with	O
knowledge	O
bases	O
,	O
which	O
is	O
beneficial	O
for	O
annotating	O
the	O
huge	O
amount	O
of	O
raw	O
and	O
often	O
noisy	O
data	O
on	O
the	O
Web	O
and	O
contributes	O
to	O
the	O
vision	O
of	O
the	O
Semantic	O
Web	O
.	O
</s>
<s>
In	O
addition	O
to	O
entity	B-General_Concept
linking	I-General_Concept
,	O
there	O
are	O
other	O
critical	O
steps	O
including	O
but	O
not	O
limited	O
to	O
event	O
extraction	O
,	O
and	O
event	O
linking	O
etc	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
is	O
beneficial	O
in	O
fields	O
that	O
need	O
to	O
extract	O
abstract	O
representations	O
from	O
text	O
,	O
as	O
it	O
happens	O
in	O
text	O
analysis	O
,	O
recommender	B-Application
systems	I-Application
,	O
semantic	O
search	O
and	O
chatbots	O
.	O
</s>
<s>
For	O
example	O
,	O
a	O
common	O
task	O
performed	O
by	O
search	B-Application
engines	I-Application
is	O
to	O
find	O
documents	O
that	O
are	O
similar	O
to	O
one	O
given	O
as	O
input	O
,	O
or	O
to	O
find	O
additional	O
information	O
about	O
the	O
persons	O
that	O
are	O
mentioned	O
in	O
it	O
.	O
</s>
<s>
Consider	O
a	O
sentence	O
that	O
contains	O
the	O
expression	O
"	O
the	O
capital	O
of	O
France	O
"	O
:	O
without	O
entity	B-General_Concept
linking	I-General_Concept
,	O
the	O
search	B-Application
engine	I-Application
that	O
looks	O
at	O
the	O
content	O
of	O
documents	O
would	O
not	O
be	O
able	O
to	O
directly	O
retrieve	O
documents	O
containing	O
the	O
word	O
"	O
Paris	O
"	O
,	O
leading	O
to	O
so-called	O
false	O
negatives	O
(	O
FN	O
)	O
.	O
</s>
<s>
Many	O
approaches	O
orthogonal	O
to	O
entity	B-General_Concept
linking	I-General_Concept
exist	O
to	O
retrieve	O
documents	O
similar	O
to	O
an	O
input	O
document	O
.	O
</s>
<s>
doc2vec	B-Algorithm
.	O
</s>
<s>
For	O
example	O
,	O
obtaining	O
schematic	O
information	O
about	O
"	O
Paris	O
"	O
,	O
as	O
presented	O
by	O
Wikipedia	O
infoboxes	O
would	O
be	O
much	O
less	O
straightforward	O
,	O
or	O
sometimes	O
even	O
unfeasible	O
,	O
depending	O
on	O
the	O
query	B-Library
complexity	O
.	O
</s>
<s>
Moreover	O
,	O
entity	B-General_Concept
linking	I-General_Concept
has	O
been	O
used	O
to	O
improve	O
the	O
performance	O
of	O
information	B-Library
retrieval	I-Library
systems	I-Library
and	O
to	O
improve	O
search	O
performance	O
on	O
digital	O
libraries	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
is	O
also	O
a	O
key	O
input	O
for	O
semantic	O
search	O
.	O
</s>
<s>
An	O
entity	B-General_Concept
linking	I-General_Concept
system	O
has	O
to	O
deal	O
with	O
a	O
number	O
of	O
challenges	O
before	O
being	O
performant	O
in	O
real-life	O
applications	O
.	O
</s>
<s>
Some	O
of	O
these	O
issues	O
are	O
intrinsic	O
to	O
the	O
task	O
of	O
entity	B-General_Concept
linking	I-General_Concept
,	O
such	O
as	O
text	O
ambiguity	O
,	O
while	O
others	O
,	O
such	O
as	O
scalability	O
and	O
execution	O
time	O
,	O
become	O
relevant	O
when	O
considering	O
real-life	O
usage	O
of	O
such	O
systems	O
.	O
</s>
<s>
Absence	O
:	O
sometimes	O
,	O
some	O
named	B-General_Concept
entities	I-General_Concept
might	O
not	O
have	O
a	O
correct	O
entity	O
link	O
in	O
the	O
target	O
knowledge	O
base	O
.	O
</s>
<s>
In	O
all	O
these	O
cases	O
,	O
the	O
entity	B-General_Concept
linking	I-General_Concept
system	O
should	O
return	O
a	O
NIL	O
entity	O
link	O
.	O
</s>
<s>
Understanding	O
when	O
to	O
return	O
a	O
NIL	O
prediction	O
is	O
not	O
straightforward	O
,	O
and	O
many	O
different	O
approaches	O
have	O
been	O
proposed	O
;	O
for	O
example	O
,	O
by	O
thresholding	O
some	O
kind	O
of	O
confidence	O
score	O
in	O
the	O
entity	B-General_Concept
linking	I-General_Concept
system	O
,	O
or	O
by	O
adding	O
an	O
additional	O
NIL	O
entity	O
to	O
the	O
knowledge	O
base	O
,	O
which	O
is	O
treated	O
in	O
the	O
same	O
way	O
as	O
the	O
other	O
entities	O
.	O
</s>
<s>
Scalability	O
and	O
Speed	O
:	O
it	O
is	O
desirable	O
for	O
an	O
industrial	O
entity	B-General_Concept
linking	I-General_Concept
system	O
to	O
provide	O
results	O
in	O
a	O
reasonable	O
time	O
,	O
and	O
often	O
in	O
real-time	O
.	O
</s>
<s>
This	O
requirement	O
is	O
critical	O
for	O
search	B-Application
engines	I-Application
,	O
chat-bots	O
and	O
for	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
offered	O
by	O
data-analytics	O
platforms	O
.	O
</s>
<s>
Evolving	O
Information	O
:	O
an	O
entity	B-General_Concept
linking	I-General_Concept
system	O
should	O
also	O
deal	O
with	O
evolving	O
information	O
,	O
and	O
easily	O
integrate	O
updates	O
in	O
the	O
knowledge	O
base	O
.	O
</s>
<s>
Multiple	O
Languages	O
:	O
an	O
entity	B-General_Concept
linking	I-General_Concept
system	O
might	O
support	O
queries	B-Library
performed	O
in	O
multiple	O
languages	O
.	O
</s>
<s>
Ideally	O
,	O
the	O
accuracy	O
of	O
the	O
entity	B-General_Concept
linking	I-General_Concept
system	O
should	O
not	O
be	O
influenced	O
by	O
the	O
input	O
language	O
,	O
and	O
entities	O
in	O
the	O
knowledge	O
base	O
should	O
be	O
the	O
same	O
across	O
different	O
languages	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
is	O
also	O
known	O
as	O
named-entity	O
disambiguation	B-General_Concept
(	O
NED	O
)	O
,	O
and	O
is	O
deeply	O
connected	O
to	O
Wikification	O
and	O
record	B-General_Concept
linkage	I-General_Concept
.	O
</s>
<s>
consider	O
entity	B-General_Concept
linking	I-General_Concept
as	O
a	O
broader	O
version	O
of	O
NED	O
,	O
as	O
NED	O
should	O
assume	O
that	O
the	O
entity	O
that	O
correctly	O
matches	O
a	O
certain	O
textual	O
named	B-General_Concept
entity	I-General_Concept
mention	O
is	O
in	O
the	O
knowledge	O
base	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
systems	O
might	O
deal	O
with	O
cases	O
in	O
which	O
no	O
entry	O
for	O
the	O
named	B-General_Concept
entity	I-General_Concept
is	O
available	O
in	O
the	O
reference	O
knowledge	O
base	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
(	O
RL	O
)	O
is	O
considered	O
a	O
broader	O
field	O
than	O
entity	B-General_Concept
linking	I-General_Concept
,	O
and	O
consists	O
in	O
finding	O
records	O
,	O
across	O
multiple	O
and	O
often	O
heterogeneous	O
data-sets	O
,	O
that	O
refer	O
to	O
the	O
same	O
entity	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
a	O
key	O
component	O
to	O
digitalize	O
archives	O
,	O
and	O
to	O
join	O
multiple	O
knowledge	O
bases	O
.	O
</s>
<s>
Named-entity	B-General_Concept
recognition	I-General_Concept
locates	O
and	O
classifies	O
named	B-General_Concept
entities	I-General_Concept
in	O
unstructured	O
text	O
into	O
pre-defined	O
categories	O
such	O
as	O
the	O
names	O
,	O
organizations	O
,	O
locations	O
,	O
and	O
more	O
.	O
</s>
<s>
Named-entity	B-General_Concept
recognition	I-General_Concept
is	O
usually	O
a	O
preprocessing	O
step	O
of	O
an	O
entity	B-General_Concept
linking	I-General_Concept
system	O
,	O
as	O
it	O
can	O
be	O
useful	O
to	O
know	O
in	O
advance	O
which	O
words	O
should	O
be	O
linked	O
to	O
entities	O
of	O
the	O
knowledge	O
base	O
.	O
</s>
<s>
A	O
notable	O
distinction	O
compared	O
to	O
entity	B-General_Concept
linking	I-General_Concept
is	O
that	O
Coreference	O
Resolution	O
does	O
not	O
assign	O
any	O
unique	O
identity	O
to	O
the	O
words	O
it	O
matches	O
,	O
but	O
it	O
simply	O
says	O
whether	O
they	O
refer	O
to	O
the	O
same	O
entity	O
or	O
not	O
.	O
</s>
<s>
In	O
that	O
sense	O
,	O
predictions	O
from	O
a	O
coreference	O
resolution	O
system	O
could	O
be	O
useful	O
to	O
a	O
subsequent	O
entity	B-General_Concept
linking	I-General_Concept
component	O
.	O
</s>
<s>
Entity	B-General_Concept
linking	I-General_Concept
has	O
been	O
a	O
hot	O
topic	O
in	O
industry	O
and	O
academia	O
for	O
the	O
last	O
decade	O
.	O
</s>
<s>
However	O
,	O
as	O
of	O
today	O
most	O
existing	O
challenges	O
are	O
still	O
unsolved	O
,	O
and	O
many	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
,	O
with	O
widely	O
different	O
strengths	O
and	O
weaknesses	O
,	O
have	O
been	O
proposed	O
.	O
</s>
<s>
Broadly	O
speaking	O
,	O
modern	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
can	O
be	O
divided	O
into	O
two	O
categories	O
:	O
</s>
<s>
Graph-based	O
approaches	O
,	O
which	O
exploit	O
the	O
structure	O
of	O
knowledge	B-Application
graphs	I-Application
to	O
represent	O
the	O
context	O
and	O
the	O
relation	O
of	O
entities	O
.	O
</s>
<s>
Often	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
cannot	O
be	O
strictly	O
categorized	O
in	O
either	O
category	O
,	O
but	O
they	O
make	O
use	O
of	O
knowledge	B-Application
graphs	I-Application
that	O
have	O
been	O
enriched	O
with	O
additional	O
textual	O
features	O
extracted	O
,	O
for	O
example	O
,	O
from	O
the	O
text	O
corpora	O
that	O
were	O
used	O
to	O
build	O
the	O
knowledge	B-Application
graphs	I-Application
themselves	O
.	O
</s>
<s>
The	O
seminal	O
work	O
by	O
Cucerzan	O
in	O
2007	O
proposed	O
one	O
of	O
the	O
first	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
that	O
appeared	O
in	O
the	O
literature	O
,	O
and	O
tackled	O
the	O
task	O
of	O
wikification	O
,	O
linking	O
textual	O
mentions	O
to	O
Wikipedia	O
pages	O
.	O
</s>
<s>
This	O
system	O
partitions	O
pages	O
as	O
entity	O
,	O
disambiguation	B-General_Concept
,	O
or	O
list	O
pages	O
,	O
used	O
to	O
assign	O
categories	O
to	O
each	O
entity	O
.	O
</s>
<s>
The	O
final	O
entity	B-General_Concept
linking	I-General_Concept
step	O
is	O
a	O
collective	O
disambiguation	B-General_Concept
performed	O
by	O
comparing	O
binary	O
vectors	O
obtained	O
from	O
hand-crafted	O
features	O
,	O
and	O
from	O
each	O
entity	O
's	O
context	O
.	O
</s>
<s>
Cucerzan	O
's	O
entity	B-General_Concept
linking	I-General_Concept
system	O
is	O
still	O
used	O
as	O
baseline	O
for	O
many	O
recent	O
works	O
.	O
</s>
<s>
is	O
a	O
well-known	O
paper	O
in	O
the	O
field	O
of	O
entity	B-General_Concept
linking	I-General_Concept
.	O
</s>
<s>
The	O
authors	O
propose	O
a	O
two-step	O
algorithm	O
to	O
link	O
named	B-General_Concept
entities	I-General_Concept
to	O
entities	O
in	O
a	O
target	O
knowledge	O
base	O
.	O
</s>
<s>
Then	O
the	O
best	O
link	O
among	O
the	O
candidates	O
is	O
chosen	O
with	O
a	O
ranking	O
support	B-Algorithm
vector	I-Algorithm
machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
that	O
uses	O
linguistic	O
features	O
.	O
</s>
<s>
Recent	O
systems	O
,	O
such	O
as	O
the	O
one	O
proposed	O
by	O
Tsai	O
et	O
al.	O
,	O
employ	O
word	B-General_Concept
embeddings	I-General_Concept
obtained	O
with	O
a	O
skip-gram	O
model	O
as	O
language	O
features	O
,	O
and	O
can	O
be	O
applied	O
to	O
any	O
language	O
as	O
long	O
as	O
a	O
large	O
corpus	O
to	O
build	O
word	B-General_Concept
embeddings	I-General_Concept
is	O
provided	O
.	O
</s>
<s>
Similarly	O
to	O
most	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
,	O
the	O
linking	O
is	O
done	O
in	O
two	O
steps	O
,	O
with	O
an	O
initial	O
candidate	O
entities	O
selection	O
and	O
a	O
linear	O
ranking	O
SVM	B-Algorithm
as	O
second	O
step	O
.	O
</s>
<s>
In	O
the	O
seminal	O
approach	O
of	O
Milne	O
and	O
Witten	O
,	O
supervised	B-General_Concept
learning	I-General_Concept
is	O
employed	O
using	O
the	O
anchor	O
texts	O
of	O
Wikipedia	O
entities	O
as	O
training	O
data	O
.	O
</s>
<s>
Modern	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
do	O
not	O
limit	O
their	O
analysis	O
to	O
textual	O
features	O
generated	O
from	O
input	O
documents	O
or	O
text	O
corpora	O
,	O
but	O
employ	O
large	O
knowledge	B-Application
graphs	I-Application
created	O
from	O
knowledge	O
bases	O
such	O
as	O
Wikipedia	O
.	O
</s>
<s>
These	O
systems	O
extract	O
complex	O
features	O
which	O
take	O
advantage	O
of	O
the	O
knowledge	B-Application
graph	I-Application
topology	O
,	O
or	O
leverage	O
multi-step	O
connections	O
between	O
entities	O
,	O
which	O
would	O
be	O
hidden	O
by	O
simple	O
text	O
analysis	O
.	O
</s>
<s>
Moreover	O
,	O
creating	O
multilingual	O
entity	B-General_Concept
linking	I-General_Concept
systems	O
based	O
on	O
natural	B-Language
language	I-Language
processing	I-Language
(	O
NLP	B-Language
)	O
is	O
inherently	O
difficult	O
,	O
as	O
it	O
requires	O
either	O
large	O
text	O
corpora	O
,	O
often	O
absent	O
for	O
many	O
languages	O
,	O
or	O
hand-crafted	O
grammar	O
rules	O
,	O
which	O
are	O
widely	O
different	O
among	O
languages	O
.	O
</s>
<s>
propose	O
the	O
creation	O
of	O
a	O
disambiguation	B-General_Concept
graph	O
(	O
a	O
subgraph	O
of	O
the	O
knowledge	O
base	O
which	O
contains	O
candidate	O
entities	O
)	O
.	O
</s>
<s>
Another	O
famous	O
entity	B-General_Concept
linking	I-General_Concept
approach	O
is	O
AIDA	O
,	O
which	O
uses	O
a	O
series	O
of	O
complex	O
graph	O
algorithms	O
,	O
and	O
a	O
greedy	O
algorithm	O
that	O
identifies	O
coherent	O
mentions	O
on	O
a	O
dense	O
subgraph	O
by	O
also	O
considering	O
context	O
similarities	O
and	O
vertex	O
importance	O
features	O
to	O
perform	O
collective	O
disambiguation	B-General_Concept
.	O
</s>
<s>
Graph	O
ranking	O
(	O
or	O
vertex	O
ranking	O
)	O
denotes	O
algorithms	O
such	O
as	O
PageRank	B-Algorithm
(	O
PR	O
)	O
and	O
Hyperlink-Induced	B-Application
Topic	I-Application
Search	I-Application
(	O
HITS	O
)	O
,	O
with	O
the	O
goal	O
to	O
assign	O
a	O
score	O
to	O
each	O
vertex	O
that	O
represents	O
its	O
relative	O
importance	O
in	O
the	O
overall	O
graph	O
.	O
</s>
<s>
The	O
entity	B-General_Concept
linking	I-General_Concept
system	O
presented	O
in	O
Alhelbawy	O
et	O
al	O
.	O
</s>
<s>
employs	O
PageRank	B-Algorithm
to	O
perform	O
collective	O
entity	B-General_Concept
linking	I-General_Concept
on	O
a	O
disambiguation	B-General_Concept
graph	O
,	O
and	O
to	O
understand	O
which	O
entities	O
are	O
more	O
strongly	O
related	O
with	O
each	O
other	O
and	O
would	O
represent	O
a	O
better	O
linking	O
.	O
</s>
<s>
This	O
is	O
essential	O
for	O
disambiguation	B-General_Concept
,	O
since	O
symbols	O
may	O
have	O
different	O
meanings	O
(	O
e.g.	O
,	O
"	O
E	O
"	O
can	O
be	O
"	O
energy	O
"	O
or	O
"	O
expectation	O
value	O
"	O
,	O
etc	O
.	O
)	O
.	O
</s>
<s>
The	O
math	O
entity	B-General_Concept
linking	I-General_Concept
process	O
can	O
be	O
facilitated	O
and	O
accelerated	O
through	O
annotation	O
recommendation	O
,	O
e.g.	O
,	O
using	O
the	O
"	O
AnnoMathTeX	O
"	O
system	O
that	O
is	O
hosted	O
by	O
Wikimedia	O
.	O
</s>
<s>
To	O
facilitate	O
the	O
reproducibility	O
of	O
Mathematical	O
Entity	B-General_Concept
Linking	I-General_Concept
(	O
MathEL	O
)	O
experiments	O
,	O
the	O
benchmark	O
MathMLben	O
was	O
created	O
.	O
</s>
<s>
describe	O
linking	O
mathematical	O
formula	O
content	O
to	O
Wikidata	O
,	O
both	O
in	O
MathML	B-Language
and	O
LaTeX	B-Application
markup	O
.	O
</s>
