<s>
Probabilistic	B-General_Concept
latent	I-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
(	O
PLSA	B-General_Concept
)	O
,	O
also	O
known	O
as	O
probabilistic	B-General_Concept
latent	I-General_Concept
semantic	I-General_Concept
indexing	I-General_Concept
(	O
PLSI	O
,	O
especially	O
in	O
information	B-Library
retrieval	I-Library
circles	O
)	O
is	O
a	O
statistical	O
technique	O
for	O
the	O
analysis	O
of	O
two-mode	O
and	O
co-occurrence	O
data	O
.	O
</s>
<s>
In	O
effect	O
,	O
one	O
can	O
derive	O
a	O
low-dimensional	O
representation	O
of	O
the	O
observed	O
variables	O
in	O
terms	O
of	O
their	O
affinity	O
to	O
certain	O
hidden	O
variables	O
,	O
just	O
as	O
in	O
latent	O
semantic	O
analysis	O
,	O
from	O
which	O
PLSA	B-General_Concept
evolved	O
.	O
</s>
<s>
Compared	O
to	O
standard	O
latent	O
semantic	O
analysis	O
which	O
stems	O
from	O
linear	B-Language
algebra	I-Language
and	O
downsizes	O
the	O
occurrence	O
tables	O
(	O
usually	O
via	O
a	O
singular	O
value	O
decomposition	O
)	O
,	O
probabilistic	B-General_Concept
latent	I-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
is	O
based	O
on	O
a	O
mixture	O
decomposition	O
derived	O
from	O
a	O
latent	O
class	O
model	O
.	O
</s>
<s>
Considering	O
observations	O
in	O
the	O
form	O
of	O
co-occurrences	O
of	O
words	O
and	O
documents	O
,	O
PLSA	B-General_Concept
models	O
the	O
probability	O
of	O
each	O
co-occurrence	O
as	O
a	O
mixture	O
of	O
conditionally	O
independent	O
multinomial	O
distributions	O
:	O
</s>
<s>
In	O
addition	O
,	O
although	O
PLSA	B-General_Concept
is	O
a	O
generative	O
model	O
of	O
the	O
documents	O
in	O
the	O
collection	O
it	O
is	O
estimated	O
on	O
,	O
it	O
is	O
not	O
a	O
generative	O
model	O
of	O
new	O
documents	O
.	O
</s>
<s>
Their	O
parameters	O
are	O
learned	O
using	O
the	O
EM	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
PLSA	B-General_Concept
may	O
be	O
used	O
in	O
a	O
discriminative	O
setting	O
,	O
via	O
Fisher	B-Algorithm
kernels	I-Algorithm
.	O
</s>
<s>
PLSA	B-General_Concept
has	O
applications	O
in	O
information	B-Library
retrieval	I-Library
and	O
filtering	B-Application
,	O
natural	B-Language
language	I-Language
processing	I-Language
,	O
machine	O
learning	O
from	O
text	O
,	O
bioinformatics	O
,	O
and	O
related	O
areas	O
.	O
</s>
<s>
It	O
is	O
reported	O
that	O
the	O
aspect	O
model	O
used	O
in	O
the	O
probabilistic	B-General_Concept
latent	I-General_Concept
semantic	I-General_Concept
analysis	I-General_Concept
has	O
severe	O
overfitting	B-Error_Name
problems	O
.	O
</s>
<s>
Generative	O
models	O
:	O
The	O
following	O
models	O
have	O
been	O
developed	O
to	O
address	O
an	O
often-criticized	O
shortcoming	O
of	O
PLSA	B-General_Concept
,	O
namely	O
that	O
it	O
is	O
not	O
a	O
proper	O
generative	O
model	O
for	O
new	O
documents	O
.	O
</s>
<s>
Higher-order	O
data	O
:	O
Although	O
this	O
is	O
rarely	O
discussed	O
in	O
the	O
scientific	O
literature	O
,	O
PLSA	B-General_Concept
extends	O
naturally	O
to	O
higher	O
order	O
data	O
(	O
three	O
modes	O
and	O
higher	O
)	O
,	O
i.e.	O
</s>
