<s>
Record	B-General_Concept
linkage	I-General_Concept
(	O
also	O
known	O
as	O
data	O
matching	O
,	O
data	O
linkage	O
,	O
entity	B-General_Concept
resolution	I-General_Concept
,	O
and	O
many	O
other	O
terms	O
)	O
is	O
the	O
task	O
of	O
finding	O
records	B-Application
in	O
a	O
data	O
set	O
that	O
refer	O
to	O
the	O
same	O
entity	O
across	O
different	O
data	O
sources	O
(	O
e.g.	O
,	O
data	O
files	O
,	O
books	O
,	O
websites	O
,	O
and	O
databases	O
)	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
necessary	O
when	O
joining	B-Language
different	O
data	O
sets	O
based	O
on	O
entities	O
that	O
may	O
or	O
may	O
not	O
share	O
a	O
common	O
identifier	O
(	O
e.g.	O
,	O
database	B-Architecture
key	I-Architecture
,	O
URI	B-Protocol
,	O
National	O
identification	O
number	O
)	O
,	O
which	O
may	O
be	O
due	O
to	O
differences	O
in	O
record	O
shape	O
,	O
storage	O
location	O
,	O
or	O
curator	O
style	O
or	O
preference	O
.	O
</s>
<s>
"	O
Record	B-General_Concept
linkage	I-General_Concept
"	O
is	O
the	O
term	O
used	O
by	O
statisticians	O
,	O
epidemiologists	O
,	O
and	O
historians	O
,	O
among	O
others	O
,	O
to	O
describe	O
the	O
process	O
of	O
joining	B-Language
records	B-Application
from	O
one	O
data	O
source	O
with	O
another	O
that	O
describe	O
the	O
same	O
entity	O
.	O
</s>
<s>
Computer	B-General_Concept
scientists	I-General_Concept
often	O
refer	O
to	O
it	O
as	O
"	O
data	O
matching	O
"	O
or	O
as	O
the	O
"	O
object	O
identity	O
problem	O
"	O
.	O
</s>
<s>
While	O
they	O
share	O
similar	O
names	O
,	O
record	B-General_Concept
linkage	I-General_Concept
and	O
Linked	B-Operating_System
Data	I-Operating_System
are	O
two	O
separate	O
approaches	O
to	O
processing	O
and	O
structuring	O
data	O
.	O
</s>
<s>
Although	O
both	O
involve	O
identifying	O
matching	O
entities	O
across	O
different	O
data	O
sets	O
,	O
record	B-General_Concept
linkage	I-General_Concept
standardly	O
equates	O
"	O
entities	O
"	O
with	O
human	O
individuals	O
;	O
by	O
contrast	O
,	O
Linked	B-Operating_System
Data	I-Operating_System
is	O
based	O
on	O
the	O
possibility	O
of	O
interlinking	O
any	O
web	O
resource	O
across	O
data	O
sets	O
,	O
using	O
a	O
correspondingly	O
broader	O
concept	O
of	O
identifier	O
,	O
namely	O
a	O
URI	B-Protocol
.	O
</s>
<s>
The	O
initial	O
idea	O
of	O
record	B-General_Concept
linkage	I-General_Concept
goes	O
back	O
to	O
Halbert	O
L	O
.	O
Dunn	O
in	O
his	O
1946	O
article	O
titled	O
"	O
Record	B-General_Concept
Linkage	I-General_Concept
"	O
published	O
in	O
the	O
American	O
Journal	O
of	O
Public	O
Health	O
.	O
</s>
<s>
Howard	O
Borden	O
Newcombe	O
then	O
laid	O
the	O
probabilistic	O
foundations	O
of	O
modern	O
record	B-General_Concept
linkage	I-General_Concept
theory	O
in	O
a	O
1959	O
article	O
in	O
Science	O
.	O
</s>
<s>
These	O
were	O
formalized	O
in	O
1969	O
by	O
Ivan	O
Fellegi	O
and	O
Alan	O
Sunter	O
,	O
in	O
their	O
pioneering	O
work	O
"	O
A	O
Theory	O
For	O
Record	B-General_Concept
Linkage	I-General_Concept
"	O
,	O
where	O
they	O
proved	O
that	O
the	O
probabilistic	O
decision	O
rule	O
they	O
described	O
was	O
optimal	O
when	O
the	O
comparison	O
attributes	O
were	O
conditionally	O
independent	O
.	O
</s>
<s>
In	O
their	O
work	O
they	O
recognized	O
the	O
growing	O
interest	O
in	O
applying	O
advances	O
in	O
computing	O
and	O
automation	O
to	O
large	O
collections	O
of	O
administrative	B-General_Concept
data	I-General_Concept
,	O
and	O
the	O
Fellegi-Sunter	B-General_Concept
theory	O
remains	O
the	O
mathematical	O
foundation	O
for	O
many	O
record	B-General_Concept
linkage	I-General_Concept
applications	O
.	O
</s>
<s>
Since	O
the	O
late	O
1990s	O
,	O
various	O
machine	O
learning	O
techniques	O
have	O
been	O
developed	O
that	O
can	O
,	O
under	O
favorable	O
conditions	O
,	O
be	O
used	O
to	O
estimate	O
the	O
conditional	O
probabilities	O
required	O
by	O
the	O
Fellegi-Sunter	B-General_Concept
theory	O
.	O
</s>
<s>
Several	O
researchers	O
have	O
reported	O
that	O
the	O
conditional	O
independence	O
assumption	O
of	O
the	O
Fellegi-Sunter	B-General_Concept
algorithm	I-General_Concept
is	O
often	O
violated	O
in	O
practice	O
;	O
however	O
,	O
published	O
efforts	O
to	O
explicitly	O
model	O
the	O
conditional	O
dependencies	O
among	O
the	O
comparison	O
attributes	O
have	O
not	O
resulted	O
in	O
an	O
improvement	O
in	O
record	B-General_Concept
linkage	I-General_Concept
quality	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
can	O
be	O
done	O
entirely	O
without	O
the	O
aid	O
of	O
a	O
computer	O
,	O
but	O
the	O
primary	O
reasons	O
computers	O
are	O
often	O
used	O
to	O
complete	O
record	B-General_Concept
linkages	I-General_Concept
are	O
to	O
reduce	O
or	O
eliminate	O
manual	O
review	O
and	O
to	O
make	O
results	O
more	O
easily	O
reproducible	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
highly	O
sensitive	O
to	O
the	O
quality	O
of	O
the	O
data	O
being	O
linked	O
,	O
so	O
all	O
data	O
sets	O
under	O
consideration	O
(	O
particularly	O
their	O
key	O
identifier	O
fields	O
)	O
should	O
ideally	O
undergo	O
a	O
data	O
quality	O
assessment	O
prior	O
to	O
record	B-General_Concept
linkage	I-General_Concept
.	O
</s>
<s>
Many	O
key	O
identifiers	O
for	O
the	O
same	O
entity	O
can	O
be	O
presented	O
quite	O
differently	O
between	O
(	O
and	O
even	O
within	O
)	O
data	O
sets	O
,	O
which	O
can	O
greatly	O
complicate	O
record	B-General_Concept
linkage	I-General_Concept
unless	O
understood	O
ahead	O
of	O
time	O
.	O
</s>
<s>
In	O
this	O
example	O
,	O
the	O
different	O
formatting	O
styles	O
lead	O
to	O
records	B-Application
that	O
look	O
different	O
but	O
in	O
fact	O
all	O
refer	O
to	O
the	O
same	O
entity	O
with	O
the	O
same	O
logical	O
identifier	O
values	O
.	O
</s>
<s>
Most	O
,	O
if	O
not	O
all	O
,	O
record	B-General_Concept
linkage	I-General_Concept
strategies	O
would	O
result	O
in	O
more	O
accurate	O
linkage	O
if	O
these	O
values	O
were	O
first	O
normalized	O
or	O
standardized	O
into	O
a	O
consistent	O
format	O
(	O
e.g.	O
,	O
all	O
names	O
are	O
"	O
Surname	O
,	O
Given	O
name	O
"	O
,	O
and	O
all	O
dates	O
are	O
"	O
YYYY/MM/DD	O
"	O
)	O
.	O
</s>
<s>
Standardization	O
can	O
be	O
accomplished	O
through	O
simple	O
rule-based	O
data	B-General_Concept
transformations	I-General_Concept
or	O
more	O
complex	O
procedures	O
such	O
as	O
lexicon-based	O
tokenization	O
and	O
probabilistic	O
hidden	O
Markov	O
models	O
.	O
</s>
<s>
Entity	B-General_Concept
resolution	I-General_Concept
is	O
an	O
operational	O
intelligence	O
process	O
,	O
typically	O
powered	O
by	O
an	O
entity	B-General_Concept
resolution	I-General_Concept
engine	O
or	O
middleware	B-General_Concept
,	O
whereby	O
organizations	O
can	O
connect	O
disparate	O
data	O
sources	O
with	O
a	O
view	O
to	O
understanding	O
possible	O
entity	O
matches	O
and	O
non-obvious	O
relationships	O
across	O
multiple	O
data	B-Application
silos	I-Application
.	O
</s>
<s>
Entity	B-General_Concept
resolution	I-General_Concept
engines	O
are	O
typically	O
used	O
to	O
uncover	O
risk	O
,	O
fraud	O
,	O
and	O
conflicts	O
of	O
interest	O
,	O
but	O
are	O
also	O
useful	O
tools	O
for	O
use	O
within	O
customer	B-General_Concept
data	I-General_Concept
integration	I-General_Concept
(	O
CDI	O
)	O
and	O
master	B-General_Concept
data	I-General_Concept
management	I-General_Concept
(	O
MDM	O
)	O
requirements	O
.	O
</s>
<s>
Typical	O
uses	O
for	O
entity	B-General_Concept
resolution	I-General_Concept
engines	O
include	O
terrorist	O
screening	O
,	O
insurance	O
fraud	O
detection	O
,	O
USA	O
Patriot	O
Act	O
compliance	O
,	O
organized	O
retail	O
crime	O
ring	O
detection	O
and	O
applicant	O
screening	O
.	O
</s>
<s>
For	O
example	O
:	O
Across	O
different	O
data	B-Application
silos	I-Application
–	O
employee	O
records	B-Application
,	O
vendor	O
data	O
,	O
watch	O
lists	O
,	O
etc	O
.	O
</s>
<s>
Entity	B-General_Concept
resolution	I-General_Concept
engines	O
then	O
apply	O
rules	O
,	O
based	O
on	O
common	O
sense	O
logic	O
,	O
to	O
identify	O
hidden	O
relationships	O
across	O
the	O
data	O
.	O
</s>
<s>
While	O
entity	B-General_Concept
resolution	I-General_Concept
solutions	O
include	O
data	O
matching	O
technology	O
,	O
many	O
data	O
matching	O
offerings	O
do	O
not	O
fit	O
the	O
definition	O
of	O
entity	B-General_Concept
resolution	I-General_Concept
.	O
</s>
<s>
Here	O
are	O
four	O
factors	O
that	O
distinguish	O
entity	B-General_Concept
resolution	I-General_Concept
from	O
data	O
matching	O
,	O
according	O
to	O
John	O
Talburt	O
,	O
director	O
of	O
the	O
UALR	O
Center	O
for	O
Advanced	O
Research	O
in	O
Entity	B-General_Concept
Resolution	I-General_Concept
and	O
Information	O
Quality	O
:	O
</s>
<s>
In	O
contrast	O
to	O
data	O
quality	O
products	O
,	O
more	O
powerful	O
identity	O
resolution	O
engines	O
also	O
include	O
a	O
rules	O
engine	O
and	O
workflow	O
process	O
,	O
which	O
apply	O
business	B-General_Concept
intelligence	I-General_Concept
to	O
the	O
resolved	O
identities	O
and	O
their	O
relationships	O
.	O
</s>
<s>
These	O
advanced	O
technologies	O
make	O
automated	B-General_Concept
decisions	I-General_Concept
and	O
impact	O
business	O
processes	O
in	O
real	O
time	O
,	O
limiting	O
the	O
need	O
for	O
human	O
intervention	O
.	O
</s>
<s>
The	O
simplest	O
kind	O
of	O
record	B-General_Concept
linkage	I-General_Concept
,	O
called	O
deterministic	O
or	O
rules-based	O
record	B-General_Concept
linkage	I-General_Concept
,	O
generates	O
links	O
based	O
on	O
the	O
number	O
of	O
individual	O
identifiers	O
that	O
match	O
among	O
the	O
available	O
data	O
sets	O
.	O
</s>
<s>
Two	O
records	B-Application
are	O
said	O
to	O
match	O
via	O
a	O
deterministic	O
record	B-General_Concept
linkage	I-General_Concept
procedure	O
if	O
all	O
or	O
some	O
identifiers	O
(	O
above	O
a	O
certain	O
threshold	O
)	O
are	O
identical	O
.	O
</s>
<s>
Deterministic	O
record	B-General_Concept
linkage	I-General_Concept
is	O
a	O
good	O
option	O
when	O
the	O
entities	O
in	O
the	O
data	O
sets	O
are	O
identified	O
by	O
a	O
common	O
identifier	O
,	O
or	O
when	O
there	O
are	O
several	O
representative	O
identifiers	O
(	O
e.g.	O
,	O
name	O
,	O
date	O
of	O
birth	O
,	O
and	O
sex	O
when	O
identifying	O
a	O
person	O
)	O
whose	O
quality	O
of	O
data	O
is	O
relatively	O
high	O
.	O
</s>
<s>
The	O
records	B-Application
in	O
two	O
data	O
sets	O
(	O
identified	O
by	O
the	O
"	O
#	O
"	O
column	O
)	O
are	O
shown	O
below	O
:	O
</s>
<s>
The	O
most	O
simple	O
deterministic	O
record	B-General_Concept
linkage	I-General_Concept
strategy	O
would	O
be	O
to	O
pick	O
a	O
single	O
identifier	O
that	O
is	O
assumed	O
to	O
be	O
uniquely	O
identifying	O
,	O
say	O
SSN	O
,	O
and	O
declare	O
that	O
records	B-Application
sharing	O
the	O
same	O
value	O
identify	O
the	O
same	O
person	O
while	O
records	B-Application
not	O
sharing	O
the	O
same	O
value	O
identify	O
different	O
people	O
.	O
</s>
<s>
Handling	O
exceptions	O
such	O
as	O
missing	O
identifiers	O
involves	O
the	O
creation	O
of	O
additional	O
record	B-General_Concept
linkage	I-General_Concept
rules	O
.	O
</s>
<s>
One	O
such	O
rule	O
in	O
the	O
case	O
of	O
missing	O
SSN	O
might	O
be	O
to	O
compare	O
name	O
,	O
date	O
of	O
birth	O
,	O
sex	O
,	O
and	O
ZIP	O
code	O
with	O
other	O
records	B-Application
in	O
hopes	O
of	O
finding	O
a	O
match	O
.	O
</s>
<s>
Running	O
names	O
through	O
a	O
phonetic	B-Algorithm
algorithm	I-Algorithm
such	O
as	O
Soundex	B-Algorithm
,	O
NYSIIS	B-Algorithm
,	O
or	O
metaphone	B-Algorithm
,	O
can	O
help	O
to	O
resolve	O
these	O
types	O
of	O
problems	O
(	O
though	O
it	O
may	O
still	O
stumble	O
over	O
surname	O
changes	O
as	O
the	O
result	O
of	O
marriage	O
or	O
divorce	O
)	O
,	O
but	O
then	O
B2	O
would	O
be	O
matched	O
only	O
with	O
A1	O
since	O
the	O
ZIP	O
code	O
in	O
A2	O
is	O
different	O
.	O
</s>
<s>
As	O
this	O
example	O
demonstrates	O
,	O
even	O
a	O
small	O
decrease	O
in	O
data	O
quality	O
or	O
small	O
increase	O
in	O
the	O
complexity	O
of	O
the	O
data	O
can	O
result	O
in	O
a	O
very	O
large	O
increase	O
in	O
the	O
number	O
of	O
rules	O
necessary	O
to	O
link	O
records	B-Application
properly	O
.	O
</s>
<s>
One	O
study	O
was	O
able	O
to	O
link	O
the	O
Social	O
Security	O
Death	O
Master	O
File	O
with	O
two	O
hospital	O
registries	O
from	O
the	O
Midwestern	O
United	O
States	O
using	O
SSN	O
,	O
NYSIIS-encoded	O
first	O
name	O
,	O
birth	O
month	O
,	O
and	O
sex	O
,	O
but	O
these	O
rules	O
may	O
not	O
work	O
as	O
well	O
with	O
data	O
sets	O
from	O
other	O
geographic	O
regions	O
or	O
with	O
data	O
collected	O
on	O
younger	O
populations	O
.	O
</s>
<s>
New	O
data	O
that	O
exhibit	O
different	O
characteristics	O
than	O
was	O
initially	O
expected	O
could	O
require	O
a	O
complete	O
rebuilding	O
of	O
the	O
record	B-General_Concept
linkage	I-General_Concept
rule	O
set	O
,	O
which	O
could	O
be	O
a	O
very	O
time-consuming	O
and	O
expensive	O
endeavor	O
.	O
</s>
<s>
Probabilistic	O
record	B-General_Concept
linkage	I-General_Concept
,	O
sometimes	O
called	O
fuzzy	O
matching	O
(	O
also	O
probabilistic	O
merging	O
or	O
fuzzy	O
merging	O
in	O
the	O
context	O
of	O
merging	O
of	O
databases	O
)	O
,	O
takes	O
a	O
different	O
approach	O
to	O
the	O
record	B-General_Concept
linkage	I-General_Concept
problem	I-General_Concept
by	O
taking	O
into	O
account	O
a	O
wider	O
range	O
of	O
potential	O
identifiers	O
,	O
computing	O
weights	O
for	O
each	O
identifier	O
based	O
on	O
its	O
estimated	O
ability	O
to	O
correctly	O
identify	O
a	O
match	O
or	O
a	O
non-match	O
,	O
and	O
using	O
these	O
weights	O
to	O
calculate	O
the	O
probability	O
that	O
two	O
given	O
records	B-Application
refer	O
to	O
the	O
same	O
entity	O
.	O
</s>
<s>
Whereas	O
deterministic	O
record	B-General_Concept
linkage	I-General_Concept
requires	O
a	O
series	O
of	O
potentially	O
complex	O
rules	O
to	O
be	O
programmed	O
ahead	O
of	O
time	O
,	O
probabilistic	O
record	B-General_Concept
linkage	I-General_Concept
methods	O
can	O
be	O
"	O
trained	O
"	O
to	O
perform	O
well	O
with	O
much	O
less	O
human	O
intervention	O
.	O
</s>
<s>
Many	O
probabilistic	O
record	B-General_Concept
linkage	I-General_Concept
algorithms	O
assign	O
match/non	O
-match	O
weights	O
to	O
identifiers	O
by	O
means	O
of	O
two	O
probabilities	O
called	O
and	O
.	O
</s>
<s>
The	O
probability	O
is	O
the	O
probability	O
that	O
an	O
identifier	O
in	O
two	O
non-matching	O
records	B-Application
will	O
agree	O
purely	O
by	O
chance	O
.	O
</s>
<s>
This	O
estimation	O
may	O
be	O
done	O
based	O
on	O
prior	O
knowledge	O
of	O
the	O
data	O
sets	O
,	O
by	O
manually	O
identifying	O
a	O
large	O
number	O
of	O
matching	O
and	O
non-matching	O
pairs	O
to	O
"	O
train	O
"	O
the	O
probabilistic	O
record	B-General_Concept
linkage	I-General_Concept
algorithm	O
,	O
or	O
by	O
iteratively	O
running	O
the	O
algorithm	O
to	O
obtain	O
closer	O
estimations	O
of	O
the	O
probability	O
.	O
</s>
<s>
Determining	O
where	O
to	O
set	O
the	O
match/non	O
-match	O
thresholds	O
is	O
a	O
balancing	O
act	O
between	O
obtaining	O
an	O
acceptable	O
sensitivity	O
(	O
or	O
recall	O
,	O
the	O
proportion	O
of	O
truly	O
matching	O
records	B-Application
that	O
are	O
linked	O
by	O
the	O
algorithm	O
)	O
and	O
positive	O
predictive	O
value	O
(	O
or	O
precision	O
,	O
the	O
proportion	O
of	O
records	B-Application
linked	O
by	O
the	O
algorithm	O
that	O
truly	O
do	O
match	O
)	O
.	O
</s>
<s>
Various	O
manual	O
and	O
automated	O
methods	O
are	O
available	O
to	O
predict	O
the	O
best	O
thresholds	O
,	O
and	O
some	O
record	B-General_Concept
linkage	I-General_Concept
software	O
packages	O
have	O
built-in	O
tools	O
to	O
help	O
the	O
user	O
find	O
the	O
most	O
acceptable	O
values	O
.	O
</s>
<s>
Blocking	O
attempts	O
to	O
restrict	O
comparisons	O
to	O
just	O
those	O
records	B-Application
for	O
which	O
one	O
or	O
more	O
particularly	O
discriminating	O
identifiers	O
agree	O
,	O
which	O
has	O
the	O
effect	O
of	O
increasing	O
the	O
positive	O
predictive	O
value	O
(	O
precision	O
)	O
at	O
the	O
expense	O
of	O
sensitivity	O
(	O
recall	O
)	O
.	O
</s>
<s>
For	O
example	O
,	O
blocking	O
based	O
on	O
a	O
phonetically	O
coded	O
surname	O
and	O
ZIP	O
code	O
would	O
reduce	O
the	O
total	O
number	O
of	O
comparisons	O
required	O
and	O
would	O
improve	O
the	O
chances	O
that	O
linked	O
records	B-Application
would	O
be	O
correct	O
(	O
since	O
two	O
identifiers	O
already	O
agree	O
)	O
,	O
but	O
would	O
potentially	O
miss	O
records	B-Application
referring	O
to	O
the	O
same	O
person	O
whose	O
surname	O
or	O
ZIP	O
code	O
was	O
different	O
(	O
due	O
to	O
marriage	O
or	O
relocation	O
,	O
for	O
instance	O
)	O
.	O
</s>
<s>
Thus	O
,	O
robust	O
record	B-General_Concept
linkage	I-General_Concept
systems	O
often	O
use	O
multiple	O
blocking	O
passes	O
to	O
group	O
data	O
in	O
various	O
ways	O
in	O
order	O
to	O
come	O
up	O
with	O
groups	O
of	O
records	B-Application
that	O
should	O
be	O
compared	O
to	O
each	O
other	O
.	O
</s>
<s>
In	O
recent	O
years	O
,	O
a	O
variety	O
of	O
machine	O
learning	O
techniques	O
have	O
been	O
used	O
in	O
record	B-General_Concept
linkage	I-General_Concept
.	O
</s>
<s>
It	O
has	O
been	O
recognized	O
that	O
the	O
classic	O
Fellegi-Sunter	B-General_Concept
algorithm	I-General_Concept
for	O
probabilistic	O
record	B-General_Concept
linkage	I-General_Concept
outlined	O
above	O
is	O
equivalent	O
to	O
the	O
Naive	B-General_Concept
Bayes	I-General_Concept
algorithm	O
in	O
the	O
field	O
of	O
machine	O
learning	O
,	O
and	O
suffers	O
from	O
the	O
same	O
assumption	O
of	O
the	O
independence	O
of	O
its	O
features	O
(	O
an	O
assumption	O
that	O
is	O
typically	O
not	O
true	O
)	O
.	O
</s>
<s>
Higher	O
accuracy	O
can	O
often	O
be	O
achieved	O
by	O
using	O
various	O
other	O
machine	O
learning	O
techniques	O
,	O
including	O
a	O
single-layer	O
perceptron	B-Algorithm
,	O
random	B-Algorithm
forest	I-Algorithm
,	O
and	O
SVM	B-Algorithm
.	O
</s>
<s>
In	O
conjunction	O
with	O
distributed	O
technologies	O
,	O
accuracy	O
and	O
scale	O
for	O
record	B-General_Concept
linkage	I-General_Concept
can	O
be	O
improved	O
further	O
.	O
</s>
<s>
High	O
quality	O
record	B-General_Concept
linkage	I-General_Concept
often	O
requires	O
a	O
human	O
–	O
machine	O
hybrid	O
system	O
to	O
safely	O
manage	O
uncertainty	O
in	O
the	O
ever	O
changing	O
streams	O
of	O
chaotic	O
big	O
data	O
.	O
</s>
<s>
Recognizing	O
that	O
linkage	O
errors	O
propagate	O
into	O
the	O
linked	B-Operating_System
data	I-Operating_System
and	O
its	O
analysis	O
,	O
interactive	O
record	B-General_Concept
linkage	I-General_Concept
systems	O
have	O
been	O
proposed	O
.	O
</s>
<s>
Interactive	O
record	B-General_Concept
linkage	I-General_Concept
is	O
defined	O
as	O
people	O
iteratively	O
fine	O
tuning	O
the	O
results	O
from	O
the	O
automated	O
methods	O
and	O
managing	O
the	O
uncertainty	O
and	O
its	O
propagation	O
to	O
subsequent	O
analyses	O
.	O
</s>
<s>
The	O
main	O
objectives	O
of	O
interactive	O
record	B-General_Concept
linkage	I-General_Concept
systems	O
is	O
to	O
manually	O
resolve	O
uncertain	O
linkages	O
and	O
validate	O
the	O
results	O
until	O
it	O
is	O
at	O
acceptable	O
levels	O
for	O
the	O
given	O
application	O
.	O
</s>
<s>
Variations	O
of	O
interactive	O
record	B-General_Concept
linkage	I-General_Concept
that	O
enhance	O
privacy	O
during	O
the	O
human	O
interaction	O
steps	O
have	O
also	O
been	O
proposed	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
increasingly	O
required	O
across	O
databases	O
held	O
by	O
different	O
organisations	O
,	O
where	O
the	O
complementary	O
data	O
held	O
by	O
these	O
organisations	O
can	O
,	O
for	O
example	O
,	O
help	O
to	O
identify	O
patients	O
that	O
are	O
susceptible	O
to	O
certain	O
adverse	O
drug	O
reactions	O
(	O
linking	O
hospital	O
,	O
doctor	O
,	O
pharmacy	O
databases	O
)	O
.	O
</s>
<s>
Privacy-preserving	O
record	B-General_Concept
linkage	I-General_Concept
(	O
PPRL	O
)	O
methods	O
have	O
been	O
developed	O
with	O
the	O
aim	O
to	O
link	O
databases	O
without	O
the	O
need	O
of	O
sharing	O
the	O
original	O
sensitive	O
values	O
between	O
the	O
organisations	O
that	O
participate	O
in	O
a	O
linkage	O
.	O
</s>
<s>
In	O
PPRL	O
,	O
generally	O
the	O
attribute	O
values	O
of	O
records	B-Application
to	O
be	O
compared	O
are	O
encoded	O
or	O
encrypted	O
in	O
some	O
form	O
.	O
</s>
<s>
A	O
popular	O
such	O
encoding	O
technique	O
used	O
are	O
Bloom	B-Algorithm
filter	I-Algorithm
,	O
which	O
allows	O
approximate	O
similarities	O
to	O
be	O
calculated	O
between	O
encoded	O
values	O
without	O
the	O
need	O
for	O
sharing	O
the	O
corresponding	O
sensitive	O
plain-text	O
values	O
.	O
</s>
<s>
The	O
techniques	O
used	O
in	O
PPRL	O
must	O
guarantee	O
that	O
no	O
participating	O
organisation	O
,	O
nor	O
any	O
external	O
adversary	O
,	O
can	O
compromise	O
the	O
privacy	O
of	O
the	O
entities	O
that	O
are	O
represented	O
by	O
records	B-Application
in	O
the	O
databases	O
being	O
linked	O
.	O
</s>
<s>
In	O
an	O
application	O
with	O
two	O
files	O
,	O
A	O
and	O
B	O
,	O
denote	O
the	O
rows	O
(	O
records	B-Application
)	O
by	O
in	O
file	O
A	O
and	O
in	O
file	O
B	O
.	O
</s>
<s>
Most	O
Master	B-General_Concept
data	I-General_Concept
management	I-General_Concept
(	O
MDM	O
)	O
products	O
use	O
a	O
record	B-General_Concept
linkage	I-General_Concept
process	O
to	O
identify	O
records	B-Application
from	O
different	O
sources	O
representing	O
the	O
same	O
real-world	O
entity	O
.	O
</s>
<s>
The	O
techniques	O
used	O
in	O
MDM	O
are	O
the	O
same	O
as	O
for	O
record	B-General_Concept
linkage	I-General_Concept
generally	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
plays	O
a	O
key	O
role	O
in	O
data	O
warehousing	O
and	O
business	B-General_Concept
intelligence	I-General_Concept
.	O
</s>
<s>
Data	O
warehouses	O
serve	O
to	O
combine	O
data	O
from	O
many	O
different	O
operational	O
source	O
systems	O
into	O
one	O
logical	B-Application
data	I-Application
model	I-Application
,	O
which	O
can	O
then	O
be	O
subsequently	O
fed	O
into	O
a	O
business	B-General_Concept
intelligence	I-General_Concept
system	I-General_Concept
for	O
reporting	O
and	O
analytics	O
.	O
</s>
<s>
Each	O
operational	O
source	O
system	O
may	O
have	O
its	O
own	O
method	O
of	O
identifying	O
the	O
same	O
entities	O
used	O
in	O
the	O
logical	B-Application
data	I-Application
model	I-Application
,	O
so	O
record	B-General_Concept
linkage	I-General_Concept
between	O
the	O
different	O
sources	O
becomes	O
necessary	O
to	O
ensure	O
that	O
the	O
information	O
about	O
a	O
particular	O
entity	O
in	O
one	O
source	O
system	O
can	O
be	O
seamlessly	O
compared	O
with	O
information	O
about	O
the	O
same	O
entity	O
from	O
another	O
source	O
system	O
.	O
</s>
<s>
Data	O
standardization	O
and	O
subsequent	O
record	B-General_Concept
linkage	I-General_Concept
often	O
occur	O
in	O
the	O
"	O
transform	O
"	O
portion	O
of	O
the	O
extract	O
,	O
transform	O
,	O
load	O
(	O
ETL	O
)	O
process	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
important	O
to	O
social	O
history	O
research	O
since	O
most	O
data	O
sets	O
,	O
such	O
as	O
census	O
records	B-Application
and	O
parish	O
registers	O
were	O
recorded	O
long	O
before	O
the	O
invention	O
of	O
National	O
identification	O
numbers	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
was	O
among	O
the	O
most	O
prominent	O
themes	O
in	O
the	O
History	O
and	O
computing	O
field	O
in	O
the	O
1980s	O
,	O
but	O
has	O
since	O
been	O
subject	O
to	O
less	O
attention	O
in	O
research	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
an	O
important	O
tool	O
in	O
creating	O
data	O
required	O
for	O
examining	O
the	O
health	O
of	O
the	O
public	O
and	O
of	O
the	O
health	O
care	O
system	O
itself	O
.	O
</s>
<s>
Data	O
sources	O
can	O
be	O
examined	O
to	O
eliminate	O
duplicate	O
records	B-Application
,	O
to	O
identify	O
under-reporting	O
and	O
missing	O
cases	O
(	O
e.g.	O
,	O
census	O
population	O
counts	O
)	O
,	O
to	O
create	O
person-oriented	O
health	O
statistics	O
,	O
and	O
to	O
generate	O
disease	O
registries	O
and	O
health	O
surveillance	O
systems	O
.	O
</s>
<s>
Record	B-General_Concept
linkage	I-General_Concept
is	O
also	O
used	O
to	O
create	O
health	O
indicators	O
.	O
</s>
<s>
If	O
infant	O
death	O
records	B-Application
are	O
matched	O
to	O
birth	O
records	B-Application
,	O
it	O
is	O
possible	O
to	O
use	O
birth	O
variables	O
,	O
such	O
as	O
birth	O
weight	O
and	O
gestational	O
age	O
,	O
along	O
with	O
mortality	O
data	O
,	O
such	O
as	O
cause	O
of	O
death	O
,	O
in	O
analyzing	O
the	O
data	O
.	O
</s>
<s>
An	O
example	O
of	O
a	O
successful	O
and	O
long-standing	O
record	B-General_Concept
linkage	I-General_Concept
system	O
allowing	O
for	O
population-based	O
medical	O
research	O
is	O
the	O
Rochester	O
Epidemiology	O
Project	O
based	O
in	O
Rochester	O
,	O
Minnesota	O
.	O
</s>
<s>
Scalability	O
:	O
Due	O
to	O
the	O
absence	O
of	O
unique	O
identifiers	O
in	O
records	B-Application
,	O
record	B-General_Concept
linkage	I-General_Concept
is	O
computationally	O
expensive	O
and	O
difficult	O
to	O
scale	O
.	O
</s>
