<s>
Named-entity	B-General_Concept
recognition	I-General_Concept
(	O
NER	O
)	O
(	O
also	O
known	O
as	O
(	O
named	O
)	O
entity	O
identification	O
,	O
entity	B-General_Concept
chunking	I-General_Concept
,	O
and	O
entity	B-General_Concept
extraction	I-General_Concept
)	O
is	O
a	O
subtask	O
of	O
information	B-General_Concept
extraction	I-General_Concept
that	O
seeks	O
to	O
locate	O
and	O
classify	O
named	B-General_Concept
entities	I-General_Concept
mentioned	O
in	O
unstructured	B-Application
text	I-Application
into	O
pre-defined	O
categories	O
such	O
as	O
person	O
names	O
,	O
organizations	O
,	O
locations	O
,	O
medical	O
codes	O
,	O
time	O
expressions	O
,	O
quantities	O
,	O
monetary	O
values	O
,	O
percentages	O
,	O
etc	O
.	O
</s>
<s>
In	O
this	O
example	O
,	O
a	O
person	O
name	O
consisting	O
of	O
one	O
token	O
,	O
a	O
two-token	O
company	O
name	O
and	O
a	O
temporal	B-General_Concept
expression	I-General_Concept
have	O
been	O
detected	O
and	O
classified	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
best	O
system	O
entering	O
MUC-7	B-General_Concept
scored	O
93.39	O
%	O
of	O
F-measure	B-General_Concept
while	O
human	O
annotators	O
scored	O
97.60	O
%	O
and	O
96.95	O
%	O
.	O
</s>
<s>
GATE	B-Language
supports	O
NER	O
across	O
many	O
languages	O
and	O
domains	O
out	O
of	O
the	O
box	O
,	O
usable	O
via	O
a	O
graphical	B-Application
interface	I-Application
and	O
a	O
Java	B-Language
API	O
.	O
</s>
<s>
OpenNLP	B-Language
includes	O
rule-based	O
and	O
statistical	O
named-entity	B-General_Concept
recognition	I-General_Concept
.	O
</s>
<s>
SpaCy	B-Application
features	O
fast	O
statistical	O
NER	O
as	O
well	O
as	O
an	O
open-source	O
named-entity	O
visualizer	O
.	O
</s>
<s>
In	O
the	O
expression	O
named	B-General_Concept
entity	I-General_Concept
,	O
the	O
word	O
named	O
restricts	O
the	O
task	O
to	O
those	O
entities	O
for	O
which	O
one	O
or	O
many	O
strings	O
,	O
such	O
as	O
words	O
or	O
phrases	O
,	O
stands	O
(	O
fairly	O
)	O
consistently	O
for	O
some	O
referent	O
.	O
</s>
<s>
Full	O
named-entity	B-General_Concept
recognition	I-General_Concept
is	O
often	O
broken	O
down	O
,	O
conceptually	O
and	O
possibly	O
also	O
in	O
implementations	O
,	O
as	O
two	O
distinct	O
problems	O
:	O
detection	O
of	O
names	O
,	O
and	O
classification	B-General_Concept
of	O
the	O
names	O
by	O
the	O
type	O
of	O
entity	O
they	O
refer	O
to	O
(	O
e.g.	O
</s>
<s>
This	O
segmentation	O
problem	O
is	O
formally	O
similar	O
to	O
chunking	B-General_Concept
.	O
</s>
<s>
The	O
second	O
phase	O
requires	O
choosing	O
an	O
ontology	B-Language
by	O
which	O
to	O
organize	O
categories	O
of	O
things	O
.	O
</s>
<s>
may	O
also	O
be	O
considered	O
as	O
named	B-General_Concept
entities	I-General_Concept
in	O
the	O
context	O
of	O
the	O
NER	O
task	O
.	O
</s>
<s>
It	O
is	O
arguable	O
that	O
the	O
definition	O
of	O
named	B-General_Concept
entity	I-General_Concept
is	O
loosened	O
in	O
such	O
cases	O
for	O
practical	O
reasons	O
.	O
</s>
<s>
The	O
definition	O
of	O
the	O
term	O
named	B-General_Concept
entity	I-General_Concept
is	O
therefore	O
not	O
strict	O
and	O
often	O
has	O
to	O
be	O
explained	O
in	O
the	O
context	O
in	O
which	O
it	O
is	O
used	O
.	O
</s>
<s>
Certain	O
hierarchies	O
of	O
named	B-General_Concept
entity	I-General_Concept
types	O
have	O
been	O
proposed	O
in	O
the	O
literature	O
.	O
</s>
<s>
BBN	O
categories	O
,	O
proposed	O
in	O
2002	O
,	O
is	O
used	O
for	O
question	B-Algorithm
answering	I-Algorithm
and	O
consists	O
of	O
29	O
types	O
and	O
64	O
subtypes	O
.	O
</s>
<s>
The	O
usual	O
measures	O
are	O
called	O
precision	O
,	O
recall	O
,	O
and	O
F1	B-General_Concept
score	I-General_Concept
.	O
</s>
<s>
In	O
academic	O
conferences	O
such	O
as	O
CoNLL	O
,	O
a	O
variant	O
of	O
the	O
F1	B-General_Concept
score	I-General_Concept
has	O
been	O
defined	O
as	O
follows	O
:	O
</s>
<s>
F1	B-General_Concept
score	I-General_Concept
is	O
the	O
harmonic	O
mean	O
of	O
these	O
two	O
.	O
</s>
<s>
Many	O
different	O
classifier	B-General_Concept
types	O
have	O
been	O
used	O
to	O
perform	O
machine-learned	O
NER	O
,	O
with	O
conditional	B-General_Concept
random	I-General_Concept
fields	I-General_Concept
being	O
a	O
typical	O
choice	O
.	O
</s>
<s>
Since	O
about	O
1998	O
,	O
there	O
has	O
been	O
a	O
great	O
deal	O
of	O
interest	O
in	O
entity	O
identification	O
in	O
the	O
molecular	O
biology	O
,	O
bioinformatics	O
,	O
and	O
medical	O
natural	B-Language
language	I-Language
processing	I-Language
communities	O
.	O
</s>
<s>
Despite	O
high	O
F1	O
numbers	O
reported	O
on	O
the	O
MUC-7	B-General_Concept
dataset	O
,	O
the	O
problem	O
of	O
named-entity	B-General_Concept
recognition	I-General_Concept
is	O
far	O
from	O
being	O
solved	O
.	O
</s>
<s>
The	O
main	O
efforts	O
are	O
directed	O
to	O
reducing	O
the	O
annotations	O
labor	O
by	O
employing	O
semi-supervised	B-General_Concept
learning	I-General_Concept
,	O
robust	O
performance	O
across	O
domains	O
and	O
scaling	O
up	O
to	O
fine-grained	O
entity	O
types	O
.	O
</s>
<s>
In	O
recent	O
years	O
,	O
many	O
projects	O
have	O
turned	O
to	O
crowdsourcing	O
,	O
which	O
is	O
a	O
promising	O
solution	O
to	O
obtain	O
high-quality	O
aggregate	O
human	O
judgments	O
for	O
supervised	B-General_Concept
and	O
semi-supervised	O
machine	O
learning	O
approaches	O
to	O
NER	O
.	O
</s>
<s>
Another	O
challenging	O
task	O
is	O
devising	O
models	O
to	O
deal	O
with	O
linguistically	O
complex	O
contexts	O
such	O
as	O
Twitter	B-Application
and	O
search	O
queries	O
.	O
</s>
<s>
There	O
are	O
some	O
researchers	O
who	O
did	O
some	O
comparisons	O
about	O
the	O
NER	O
performances	O
from	O
different	O
statistical	O
models	O
such	O
as	O
HMM	O
(	O
hidden	O
Markov	O
model	O
)	O
,	O
ME	O
(	O
maximum	O
entropy	O
)	O
,	O
and	O
CRF	O
(	O
conditional	B-General_Concept
random	I-General_Concept
fields	I-General_Concept
)	O
,	O
and	O
feature	O
sets	O
.	O
</s>
<s>
And	O
some	O
researchers	O
recently	O
proposed	O
graph-based	O
semi-supervised	B-General_Concept
learning	I-General_Concept
model	O
for	O
language	O
specific	O
NER	O
tasks	O
.	O
</s>
<s>
A	O
recently	O
emerging	O
task	O
of	O
identifying	O
"	O
important	O
expressions	O
"	O
in	O
text	O
and	O
cross-linking	B-General_Concept
them	I-General_Concept
to	I-General_Concept
Wikipedia	I-General_Concept
can	O
be	O
seen	O
as	O
an	O
instance	O
of	O
extremely	O
fine-grained	O
named-entity	B-General_Concept
recognition	I-General_Concept
,	O
where	O
the	O
types	O
are	O
the	O
actual	O
Wikipedia	O
pages	O
describing	O
the	O
(	O
potentially	O
ambiguous	O
)	O
concepts	O
.	O
</s>
<s>
Another	O
field	O
that	O
has	O
seen	O
progress	O
but	O
remains	O
challenging	O
is	O
the	O
application	O
of	O
NER	O
to	O
Twitter	B-Application
and	O
other	O
microblogs	O
,	O
considered	O
"	O
noisy	O
"	O
due	O
to	O
non-standard	O
orthography	O
,	O
shortness	O
and	O
informality	O
of	O
texts	O
.	O
</s>
<s>
NER	O
challenges	O
in	O
English	O
Tweets	O
have	O
been	O
organized	O
by	O
research	O
communities	O
to	O
compare	O
performances	O
of	O
various	O
approaches	O
,	O
such	O
as	O
bidirectional	B-Algorithm
LSTMs	I-Algorithm
,	O
Learning-to-Search	O
,	O
or	O
CRFs	O
.	O
</s>
