<s>
In	O
natural	B-Language
language	I-Language
processing	I-Language
and	O
information	B-Library
retrieval	I-Library
,	O
explicit	B-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
(	O
ESA	O
)	O
is	O
a	O
vectoral	O
representation	O
of	O
text	O
(	O
individual	O
words	O
or	O
entire	O
documents	O
)	O
that	O
uses	O
a	O
document	O
corpus	O
as	O
a	O
knowledge	O
base	O
.	O
</s>
<s>
Typically	O
,	O
the	O
text	O
corpus	O
is	O
English	O
Wikipedia	O
,	O
though	O
other	O
corpora	O
including	O
the	B-Application
Open	I-Application
Directory	I-Application
Project	I-Application
have	O
been	O
used	O
.	O
</s>
<s>
and	O
has	O
been	O
used	O
by	O
this	O
pair	O
of	O
researchers	O
to	O
compute	O
what	O
they	O
refer	O
to	O
as	O
"	O
semantic	B-Application
relatedness	O
"	O
by	O
means	O
of	O
cosine	O
similarity	O
between	O
the	O
aforementioned	O
vectors	O
,	O
collectively	O
interpreted	O
as	O
a	O
space	O
of	O
"	O
concepts	O
explicitly	O
defined	O
and	O
described	O
by	O
humans	O
"	O
,	O
where	O
Wikipedia	O
articles	O
(	O
or	O
ODP	O
entries	O
,	O
or	O
otherwise	O
titles	O
of	O
documents	O
in	O
the	O
knowledge	O
base	O
corpus	O
)	O
are	O
equated	O
with	O
concepts	O
.	O
</s>
<s>
The	O
name	O
"	O
explicit	B-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
"	O
contrasts	O
with	O
latent	O
semantic	B-Application
analysis	O
(	O
LSA	O
)	O
,	O
because	O
the	O
use	O
of	O
a	O
knowledge	O
base	O
makes	O
it	O
possible	O
to	O
assign	O
human-readable	O
labels	O
to	O
the	O
concepts	O
that	O
make	O
up	O
the	O
vector	O
space	O
.	O
</s>
<s>
These	O
are	O
all	O
turned	O
into	O
"	O
bags	B-General_Concept
of	I-General_Concept
words	I-General_Concept
"	O
,	O
i.e.	O
,	O
term	O
frequency	O
histograms	O
,	O
stored	O
in	O
an	O
inverted	B-Algorithm
index	I-Algorithm
.	O
</s>
<s>
Using	O
this	O
inverted	B-Algorithm
index	I-Algorithm
,	O
one	O
can	O
find	O
for	O
any	O
word	O
the	O
set	O
of	O
Wikipedia	O
articles	O
containing	O
this	O
word	O
;	O
in	O
the	O
vocabulary	O
of	O
Egozi	O
,	O
Markovitch	O
and	O
Gabrilovitch	O
,	O
"	O
each	O
word	O
appearing	O
in	O
the	O
Wikipedia	O
corpus	O
can	O
be	O
seen	O
as	O
triggering	O
each	O
of	O
the	O
concepts	O
it	O
points	O
to	O
in	O
the	O
inverted	O
index.	O
"	O
</s>
<s>
The	O
output	O
of	O
the	O
inverted	B-Algorithm
index	I-Algorithm
for	O
a	O
single	O
word	O
query	B-Library
is	O
a	O
list	O
of	O
indexed	O
documents	O
(	O
Wikipedia	O
articles	O
)	O
,	O
each	O
given	O
a	O
score	O
depending	O
on	O
how	O
often	O
the	O
word	O
in	O
question	O
occurred	O
in	O
them	O
(	O
weighted	O
by	O
the	O
total	O
number	O
of	O
words	O
in	O
the	O
document	O
)	O
.	O
</s>
<s>
Mathematically	O
,	O
this	O
list	O
is	O
an	O
-dimensional	O
vector	O
of	O
word-document	O
scores	O
,	O
where	O
a	O
document	O
not	O
containing	O
the	O
query	B-Library
word	O
has	O
score	O
zero	O
.	O
</s>
<s>
and	O
this	O
gives	O
a	O
numeric	O
estimate	O
of	O
the	O
semantic	B-Application
relatedness	O
of	O
the	O
words	O
.	O
</s>
<s>
ESA	O
,	O
as	O
originally	O
posited	O
by	O
Gabrilovich	O
and	O
Markovitch	O
,	O
operates	O
under	O
the	O
assumption	O
that	O
the	O
knowledge	O
base	O
contains	O
topically	O
orthogonal	B-Application
concepts	O
.	O
</s>
<s>
However	O
,	O
it	O
was	O
later	O
shown	O
by	O
Anderka	O
and	O
Stein	O
that	O
ESA	O
also	O
improves	O
the	O
performance	O
of	O
information	B-Library
retrieval	I-Library
systems	I-Library
when	O
it	O
is	O
based	O
not	O
on	O
Wikipedia	O
,	O
but	O
on	O
the	O
Reuters	O
corpus	O
of	O
newswire	O
articles	O
,	O
which	O
does	O
not	O
satisfy	O
the	O
orthogonality	B-Application
property	O
;	O
in	O
their	O
experiments	O
,	O
Anderka	O
and	O
Stein	O
used	O
newswire	O
stories	O
as	O
"	O
concepts	O
"	O
.	O
</s>
<s>
ESA	O
is	O
considered	O
by	O
its	O
authors	O
a	O
measure	O
of	O
semantic	B-Application
relatedness	O
(	O
as	O
opposed	O
to	O
semantic	B-Application
similarity	O
)	O
.	O
</s>
<s>
On	O
datasets	O
used	O
to	O
benchmark	O
relatedness	O
of	O
words	O
,	O
ESA	O
outperforms	O
other	O
algorithms	O
,	O
including	O
WordNet	B-General_Concept
semantic	B-Application
similarity	O
measures	O
and	O
skip-gram	O
Neural	O
Network	O
Language	O
Model	O
(	O
Word2vec	B-Algorithm
)	O
.	O
</s>
<s>
Cross-language	O
explicit	B-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
(	O
CL-ESA	O
)	O
is	O
a	O
multilingual	O
generalization	O
of	O
ESA	O
.	O
</s>
