<s>
A	O
document-term	B-General_Concept
matrix	I-General_Concept
is	O
a	O
mathematical	O
matrix	B-Architecture
that	O
describes	O
the	O
frequency	O
of	O
terms	O
that	O
occur	O
in	O
a	O
collection	O
of	O
documents	O
.	O
</s>
<s>
In	O
a	O
document-term	B-General_Concept
matrix	I-General_Concept
,	O
rows	O
correspond	O
to	O
documents	O
in	O
the	O
collection	O
and	O
columns	O
correspond	O
to	O
terms	O
.	O
</s>
<s>
This	O
matrix	B-Architecture
is	O
a	O
specific	O
instance	O
of	O
a	O
document-feature	O
matrix	B-Architecture
where	O
"	O
features	O
"	O
may	O
refer	O
to	O
other	O
properties	O
of	O
a	O
document	O
besides	O
terms	O
.	O
</s>
<s>
It	O
is	O
also	O
common	O
to	O
encounter	O
the	O
transpose	O
,	O
or	O
term-document	B-General_Concept
matrix	I-General_Concept
where	O
documents	O
are	O
the	O
columns	O
and	O
terms	O
are	O
the	O
rows	O
.	O
</s>
<s>
They	O
are	O
useful	O
in	O
the	O
field	O
of	O
natural	B-Language
language	I-Language
processing	I-Language
and	O
computational	O
text	O
analysis	O
.	O
</s>
<s>
In	O
such	O
a	O
case	O
,	O
this	O
is	O
also	O
referred	O
to	O
as	O
"	O
bag	B-General_Concept
of	I-General_Concept
words	I-General_Concept
"	O
representation	O
because	O
the	O
counts	O
of	O
individual	O
words	O
is	O
retained	O
,	O
but	O
not	O
the	O
order	O
of	O
the	O
words	O
in	O
the	O
document	O
.	O
</s>
<s>
When	O
creating	O
a	O
data-set	O
of	O
terms	O
that	O
appear	O
in	O
a	O
corpus	O
of	O
documents	O
,	O
the	O
document-term	B-General_Concept
matrix	I-General_Concept
contains	O
rows	O
corresponding	O
to	O
the	O
documents	O
and	O
columns	O
corresponding	O
to	O
the	O
terms	O
.	O
</s>
<s>
then	O
the	O
document-term	B-General_Concept
matrix	I-General_Concept
would	O
be	O
:	O
</s>
<s>
Note	O
that	O
,	O
unlike	O
representing	O
a	O
document	O
as	O
just	O
a	O
token-count	O
list	O
,	O
the	O
document-term	B-General_Concept
matrix	I-General_Concept
includes	O
all	O
terms	O
in	O
the	O
corpus	O
(	O
i.e.	O
</s>
<s>
The	O
document-term	B-General_Concept
matrix	I-General_Concept
emerged	O
in	O
the	O
earliest	O
years	O
of	O
the	O
computerization	O
of	O
text	O
.	O
</s>
<s>
Certain	O
function	O
words	O
such	O
as	O
and	O
,	O
the	O
,	O
at	O
,	O
a	O
,	O
etc.	O
,	O
were	O
placed	O
in	O
a	O
"	O
forbidden	O
word	O
list	O
"	O
table	O
,	O
and	O
the	O
frequency	O
of	O
these	O
words	O
was	O
recorded	O
in	O
a	O
separate	O
listing	O
...	O
A	O
special	O
computer	O
program	O
,	O
called	O
the	O
Descriptor	O
Word	O
Index	O
Program	O
,	O
was	O
written	O
to	O
provide	O
this	O
information	O
and	O
to	O
prepare	O
a	O
document-term	B-General_Concept
matrix	I-General_Concept
in	O
a	O
form	O
suitable	O
for	O
in-put	O
to	O
the	O
Factor	O
Analysis	O
Program	O
.	O
</s>
<s>
The	O
Descriptor	O
Word	O
Index	O
program	O
was	O
prepared	O
by	O
Eileen	O
Stone	O
of	O
the	O
System	O
Development	O
Corporation.Shortly	O
thereafter	O
,	O
Gerard	O
Salton	O
published	O
"	O
Some	O
hierarchical	O
models	O
for	O
automatic	O
document	O
retrieval	O
"	O
in	O
1963	O
which	O
also	O
included	O
a	O
visual	O
depiction	O
of	O
a	O
document-term	B-General_Concept
matrix	I-General_Concept
.	O
</s>
<s>
In	O
this	O
paper	O
,	O
Salton	O
introduces	O
the	O
document-term	B-General_Concept
matrix	I-General_Concept
by	O
comparison	O
to	O
a	O
kind	O
of	O
term-context	O
matrix	B-Architecture
used	O
to	O
measure	O
similarities	O
between	O
words:If	O
it	O
is	O
desired	O
to	O
generate	O
document	O
associations	O
or	O
document	O
clusters	O
instead	O
of	O
word	O
associations	O
,	O
the	O
same	O
procedures	O
can	O
be	O
used	O
with	O
slight	O
modifications	O
.	O
</s>
<s>
Instead	O
of	O
starting	O
with	O
a	O
word-sentence	O
matrix	B-Architecture
C	O
,...	O
it	O
is	O
now	O
convenient	O
to	O
construct	O
a	O
word-document	O
matrix	B-Architecture
F	O
,	O
listing	O
frequency	O
of	O
occurrence	O
of	O
word	O
Wi	O
in	O
Document	O
Dj	O
...	O
</s>
<s>
This	O
procedure	O
produces	O
a	O
document-document	O
similarity	O
matrix	B-Architecture
which	O
can	O
in	O
turn	O
be	O
used	O
for	O
the	O
generation	O
of	O
document	O
clusters	O
...	O
In	O
addition	O
to	O
Borko	O
and	O
Salton	O
,	O
in	O
1964	O
,	O
F.W.	O
</s>
<s>
These	O
are	O
arranged	O
in	O
a	O
document/term	O
matrix	B-Architecture
to	O
show	O
the	O
frequency	O
of	O
occurrence	O
of	O
each	O
term	O
in	O
each	O
document	O
....	O
A	O
correlation	O
coefficient	O
for	O
each	O
word	O
pair	O
is	O
then	O
computed	O
,	O
based	O
on	O
their	O
co-occurrence	O
in	O
the	O
document	O
set	O
.	O
</s>
<s>
The	O
resulting	O
term/term	O
matrix	B-Architecture
...	O
is	O
then	O
factor	O
analysed	O
and	O
a	O
series	O
of	O
factors	O
are	O
isolated	O
.	O
</s>
<s>
A	O
point	O
of	O
view	O
on	O
the	O
matrix	B-Architecture
is	O
that	O
each	O
row	O
represents	O
a	O
document	O
.	O
</s>
<s>
In	O
the	O
vectorial	O
semantic	O
model	O
,	O
which	O
is	O
normally	O
the	O
one	O
used	O
to	O
compute	O
a	O
document-term	B-General_Concept
matrix	I-General_Concept
,	O
the	O
goal	O
is	O
to	O
represent	O
the	O
topic	O
of	O
a	O
document	O
by	O
the	O
frequency	O
of	O
semantically	O
significant	O
terms	O
.	O
</s>
<s>
Adding	O
collocation	B-General_Concept
as	O
terms	O
improves	O
the	O
quality	O
of	O
the	O
vectors	O
,	O
especially	O
when	O
computing	O
similarities	O
between	O
documents	O
.	O
</s>
<s>
Latent	O
semantic	O
analysis	O
(	O
LSA	O
,	O
performing	O
singular-value	O
decomposition	O
on	O
the	O
document-term	B-General_Concept
matrix	I-General_Concept
)	O
can	O
improve	O
search	O
results	O
by	O
disambiguating	B-General_Concept
polysemous	O
words	O
and	O
searching	O
for	O
synonyms	B-Application
of	O
the	O
query	O
.	O
</s>
<s>
However	O
,	O
searching	O
in	O
the	O
high-dimensional	O
continuous	O
space	O
is	O
much	O
slower	O
than	O
searching	O
the	O
standard	O
trie	B-General_Concept
data	O
structure	O
of	O
search	O
engines	O
.	O
</s>
<s>
Multivariate	O
analysis	O
of	O
the	O
document-term	B-General_Concept
matrix	I-General_Concept
can	O
reveal	O
topics/themes	O
of	O
the	O
corpus	O
.	O
</s>
<s>
Specifically	O
,	O
latent	O
semantic	O
analysis	O
and	O
data	B-Algorithm
clustering	I-Algorithm
can	O
be	O
used	O
,	O
and	O
,	O
more	O
recently	O
,	O
probabilistic	B-General_Concept
latent	I-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
with	O
its	O
generalization	O
Latent	O
Dirichlet	O
allocation	O
,	O
and	O
non-negative	O
matrix	B-Architecture
factorization	O
,	O
have	O
been	O
found	O
to	O
perform	O
well	O
for	O
this	O
task	O
.	O
</s>
