<s>
In	O
machine	O
learning	O
and	O
information	B-Library
retrieval	I-Library
,	O
the	O
cluster	B-Application
hypothesis	I-Application
is	O
an	O
assumption	O
about	O
the	O
nature	O
of	O
the	O
data	O
handled	O
in	O
those	O
fields	O
,	O
which	O
takes	O
various	O
forms	O
.	O
</s>
<s>
In	O
information	B-Library
retrieval	I-Library
,	O
it	O
states	O
that	O
documents	O
that	O
are	O
clustered	B-Algorithm
together	O
"	O
behave	O
similarly	O
with	O
respect	O
to	O
relevance	O
to	O
information	O
needs	O
"	O
.	O
</s>
<s>
In	O
terms	O
of	O
classification	B-General_Concept
,	O
it	O
states	O
that	O
if	O
points	O
are	O
in	O
the	O
same	O
cluster	O
,	O
they	O
are	O
likely	O
to	O
be	O
of	O
the	O
same	O
class	O
.	O
</s>
<s>
Thus	O
,	O
theoretically	O
,	O
a	O
search	B-Application
engine	I-Application
could	O
try	O
to	O
locate	O
only	O
the	O
appropriate	O
cluster	O
for	O
a	O
query	B-Library
,	O
and	O
then	O
allow	O
users	O
to	O
browse	O
through	O
this	O
cluster	O
.	O
</s>
<s>
Although	O
experiments	O
showed	O
that	O
the	O
cluster	B-Application
hypothesis	I-Application
as	O
such	O
holds	O
,	O
exploiting	O
it	O
for	O
retrieval	O
did	O
not	O
lead	O
to	O
satisfying	O
results	O
.	O
</s>
<s>
The	O
cluster	B-Application
assumption	I-Application
is	O
assumed	O
in	O
many	O
machine	O
learning	O
algorithms	O
such	O
as	O
the	O
k-nearest	B-General_Concept
neighbor	I-General_Concept
classification	B-General_Concept
algorithm	O
and	O
the	O
k-means	B-Algorithm
clustering	I-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
The	O
cluster	B-Application
assumption	I-Application
is	O
equivalent	O
to	O
the	O
Low	O
density	O
separation	O
assumption	O
which	O
states	O
that	O
the	O
decision	O
boundary	O
should	O
lie	O
on	O
a	O
low-density	O
region	O
.	O
</s>
