<s>
Sentence	B-General_Concept
extraction	I-General_Concept
is	O
a	O
technique	O
used	O
for	O
automatic	B-Application
summarization	I-Application
of	O
a	O
text	O
.	O
</s>
<s>
In	O
this	O
shallow	O
approach	O
,	O
statistical	B-Algorithm
heuristics	I-Algorithm
are	O
used	O
to	O
identify	O
the	O
most	O
salient	O
sentences	O
of	O
a	O
text	O
.	O
</s>
<s>
Sentence	B-General_Concept
extraction	I-General_Concept
is	O
a	O
low-cost	O
approach	O
compared	O
to	O
more	O
knowledge-intensive	O
deeper	O
approaches	O
which	O
require	O
additional	O
knowledge	O
bases	O
such	O
as	O
ontologies	O
or	O
linguistic	O
knowledge	O
.	O
</s>
<s>
In	O
short	O
"	O
sentence	B-General_Concept
extraction	I-General_Concept
"	O
works	O
as	O
a	O
filter	O
which	O
allows	O
only	O
important	O
sentences	O
to	O
pass	O
.	O
</s>
<s>
Nevertheless	O
,	O
sentence	B-General_Concept
extraction	I-General_Concept
summaries	O
can	O
give	O
valuable	O
clues	O
to	O
the	O
main	O
points	O
of	O
a	O
document	O
and	O
are	O
frequently	O
sufficiently	O
intelligible	O
to	O
human	O
readers	O
.	O
</s>
<s>
Usually	O
,	O
a	O
combination	O
of	O
heuristics	B-Algorithm
is	O
used	O
to	O
determine	O
the	O
most	O
important	O
sentences	O
within	O
the	O
document	O
.	O
</s>
<s>
Each	O
heuristic	B-Algorithm
assigns	O
a	O
(	O
positive	O
or	O
negative	O
)	O
score	O
to	O
the	O
sentence	O
.	O
</s>
<s>
After	O
all	O
heuristics	B-Algorithm
have	O
been	O
applied	O
,	O
the	O
highest-scoring	O
sentences	O
are	O
included	O
in	O
the	O
summary	O
.	O
</s>
<s>
The	O
individual	O
heuristics	B-Algorithm
are	O
weighted	O
according	O
to	O
their	O
importance	O
.	O
</s>
<s>
words	O
which	O
occur	O
significantly	O
frequently	O
in	O
the	O
document	O
,	O
is	O
still	O
one	O
of	O
the	O
core	O
heuristics	B-Algorithm
of	O
today	O
's	O
summarizers	O
.	O
</s>
<s>
With	O
large	O
linguistic	O
corpora	O
available	O
today	O
,	O
the	O
tf	O
–	O
idf	O
value	O
which	O
originated	O
in	O
information	B-Library
retrieval	I-Library
,	O
can	O
be	O
successfully	O
applied	O
to	O
identify	O
the	O
key	O
words	O
of	O
a	O
text	O
:	O
If	O
for	O
example	O
the	O
word	O
"	O
cat	O
"	O
occurs	O
significantly	O
more	O
often	O
in	O
the	O
text	O
to	O
be	O
summarized	O
(	O
TF	O
=	O
"	O
term	O
frequency	O
"	O
)	O
than	O
in	O
the	O
corpus	O
(	O
IDF	O
means	O
"	O
inverse	O
document	O
frequency	O
"	O
;	O
here	O
the	O
corpus	O
is	O
meant	O
by	O
"	O
document	O
"	O
)	O
,	O
then	O
"	O
cat	O
"	O
is	O
likely	O
to	O
be	O
an	O
important	O
word	O
of	O
the	O
text	O
;	O
the	O
text	O
may	O
in	O
fact	O
be	O
a	O
text	O
about	O
cats	O
.	O
</s>
