<s>
In	O
statistical	O
analysis	O
,	O
change	B-General_Concept
detection	I-General_Concept
or	O
change	B-General_Concept
point	I-General_Concept
detection	I-General_Concept
tries	O
to	O
identify	O
times	O
when	O
the	O
probability	O
distribution	O
of	O
a	O
stochastic	O
process	O
or	O
time	O
series	O
changes	O
.	O
</s>
<s>
Specific	O
applications	O
,	O
like	O
step	B-Algorithm
detection	I-Algorithm
and	O
edge	B-Algorithm
detection	I-Algorithm
,	O
may	O
be	O
concerned	O
with	O
changes	O
in	O
the	O
mean	O
,	O
variance	O
,	O
correlation	O
,	O
or	O
spectral	B-Algorithm
density	I-Algorithm
of	O
the	O
process	O
.	O
</s>
<s>
More	O
generally	O
change	B-General_Concept
detection	I-General_Concept
also	O
includes	O
the	O
detection	O
of	O
anomalous	O
behavior	O
:	O
anomaly	B-Algorithm
detection	I-Algorithm
.	O
</s>
<s>
Change	B-General_Concept
point	I-General_Concept
detection	I-General_Concept
is	O
concerned	O
with	O
identifying	O
whether	O
,	O
and	O
if	O
so	O
when	O
,	O
the	O
behavior	O
of	O
the	O
series	O
changes	O
significantly	O
.	O
</s>
<s>
The	O
problem	O
of	O
change	B-General_Concept
point	I-General_Concept
detection	I-General_Concept
can	O
be	O
narrowed	O
down	O
further	O
into	O
more	O
specific	O
problems	O
.	O
</s>
<s>
In	O
offline	B-Algorithm
change	B-General_Concept
point	I-General_Concept
detection	I-General_Concept
it	O
is	O
assumed	O
that	O
a	O
sequence	O
of	O
length	O
is	O
available	O
and	O
the	O
goal	O
is	O
to	O
identify	O
whether	O
any	O
change	O
point(s )	O
occurred	O
in	O
the	O
series	O
.	O
</s>
<s>
By	O
contrast	O
,	O
online	O
change	B-General_Concept
point	I-General_Concept
detection	I-General_Concept
is	O
concerned	O
with	O
detecting	O
change	O
points	O
in	O
an	O
incoming	O
data	O
stream	O
.	O
</s>
<s>
Online	O
change	B-General_Concept
detection	I-General_Concept
is	O
also	O
done	O
using	O
streaming	O
algorithms	O
.	O
</s>
<s>
Basseville	O
(	O
1993	O
,	O
Section	O
2.6	O
)	O
discusses	O
offline	B-Algorithm
change-in-mean	O
detection	O
with	O
hypothesis	O
testing	O
based	O
on	O
the	O
works	O
of	O
Page	O
and	O
Picard	O
and	O
maximum-likelihood	O
estimation	O
of	O
the	O
change	O
time	O
,	O
related	O
to	O
two-phase	O
regression	O
.	O
</s>
<s>
Other	O
approaches	O
employ	O
clustering	B-Algorithm
based	O
on	O
maximum	O
likelihood	O
estimation	O
,,	O
use	O
optimization	O
to	O
infer	O
the	O
number	O
and	O
times	O
of	O
changes	O
,	O
via	O
spectral	O
analysis	O
,	O
or	O
singular	O
spectrum	O
analysis	O
.	O
</s>
<s>
"	O
Offline	B-Algorithm
"	O
approaches	O
cannot	O
be	O
used	O
on	O
streaming	O
data	O
because	O
they	O
need	O
to	O
compare	O
to	O
statistics	O
of	O
the	O
complete	O
time	O
series	O
,	O
and	O
cannot	O
react	O
to	O
changes	O
in	O
real-time	O
but	O
often	O
provide	O
a	O
more	O
accurate	O
estimation	O
of	O
the	O
change	O
time	O
and	O
magnitude	O
.	O
</s>
<s>
Change	B-General_Concept
detection	I-General_Concept
tests	O
are	O
often	O
used	O
in	O
manufacturing	O
(	O
quality	O
control	O
)	O
,	O
intrusion	O
detection	O
,	O
spam	O
filtering	O
,	O
website	O
tracking	O
,	O
and	O
medical	O
diagnostics	O
.	O
</s>
<s>
Linguistic	O
change	B-General_Concept
detection	I-General_Concept
refers	O
to	O
the	O
ability	O
to	O
detect	O
word-level	O
changes	O
across	O
multiple	O
presentations	O
of	O
the	O
same	O
sentence	O
.	O
</s>
<s>
Researchers	O
have	O
found	O
that	O
the	O
amount	O
of	O
semantic	B-Application
overlap	O
(	O
i.e.	O
,	O
relatedness	O
)	O
between	O
the	O
changed	O
word	O
and	O
the	O
new	O
word	O
influences	O
the	O
ease	O
with	O
which	O
such	O
a	O
detection	O
is	O
made	O
(	O
Sturt	O
,	O
Sanford	O
,	O
Stewart	O
,	O
&	O
Dawydiak	O
,	O
2004	O
)	O
.	O
</s>
<s>
Recently	O
,	O
researchers	O
have	O
detected	O
word-level	O
changes	O
in	O
semantics	B-Application
across	O
time	O
by	O
computationally	O
analyzing	O
temporal	O
corpora	O
(	O
for	O
example:the	O
word	O
"	O
gay	O
"	O
has	O
acquired	O
a	O
new	O
meaning	O
over	O
time	O
)	O
using	O
change	B-General_Concept
point	I-General_Concept
detection	I-General_Concept
.	O
</s>
