<s>
The	O
Text	B-Language
REtrieval	I-Language
Conference	I-Language
(	O
TREC	O
)	O
is	O
an	O
ongoing	O
series	O
of	O
workshops	O
focusing	O
on	O
a	O
list	O
of	O
different	O
information	B-Library
retrieval	I-Library
(	O
IR	B-Library
)	O
research	O
areas	O
,	O
or	O
tracks	O
.	O
</s>
<s>
Its	O
purpose	O
is	O
to	O
support	O
and	O
encourage	O
research	O
within	O
the	O
information	B-Library
retrieval	I-Library
community	O
by	O
providing	O
the	O
infrastructure	O
necessary	O
for	O
large-scale	O
evaluation	O
of	O
text	B-Algorithm
retrieval	I-Algorithm
methodologies	O
and	O
to	O
increase	O
the	O
speed	O
of	O
lab-to-product	O
transfer	O
of	O
technology	O
.	O
</s>
<s>
A	O
2010	O
study	O
estimated	O
that	O
"	O
without	O
TREC	O
,	O
U.S.	O
Internet	O
users	O
would	O
have	O
spent	O
up	O
to	O
3.15	O
billion	O
additional	O
hours	O
using	O
web	B-Application
search	I-Application
engines	I-Application
between	O
1999	O
and	O
2009.	O
"	O
</s>
<s>
Hal	O
Varian	O
the	O
Chief	O
Economist	O
at	O
Google	B-Application
wrote	O
that	O
"	O
The	O
TREC	O
data	O
revitalized	O
research	O
on	O
information	B-Library
retrieval	I-Library
.	O
</s>
<s>
Depending	O
on	O
track	O
,	O
test	O
problems	O
might	O
be	O
questions	O
,	O
topics	O
,	O
or	O
target	O
extractable	O
features	B-Algorithm
.	O
</s>
<s>
Its	O
purpose	O
was	O
to	O
support	O
research	O
within	O
the	O
information	B-Library
retrieval	I-Library
community	O
by	O
providing	O
the	O
infrastructure	O
necessary	O
for	O
large-scale	O
evaluation	O
of	O
text	B-Algorithm
retrieval	I-Algorithm
methodologies	O
.	O
</s>
<s>
Since	O
size	O
of	O
TREC	O
collection	O
is	O
large	O
,	O
it	O
is	O
impossible	O
to	O
calculate	O
the	O
absolute	O
recall	O
for	O
each	O
query	B-Library
.	O
</s>
<s>
In	O
order	O
to	O
assess	O
the	O
relevance	O
of	O
documents	O
in	O
relation	O
to	O
a	O
query	B-Library
,	O
TREC	O
uses	O
a	O
specific	O
method	O
call	O
pooling	O
for	O
calculating	O
relative	O
recall	O
.	O
</s>
<s>
All	O
the	O
relevant	O
documents	O
that	O
occurred	O
in	O
the	O
top	O
100	O
documents	O
for	O
each	O
system	O
and	O
for	O
each	O
query	B-Library
are	O
combined	O
to	O
produce	O
a	O
pool	O
of	O
relevant	O
documents	O
.	O
</s>
<s>
Recall	O
being	O
the	O
proportion	O
of	O
the	O
pool	O
of	O
relevant	O
documents	O
that	O
a	O
single	O
system	O
retrieved	O
for	O
a	O
query	B-Library
topic	O
.	O
</s>
<s>
It	O
demonstrated	O
a	O
wide	O
range	O
of	O
different	O
approaches	O
to	O
the	O
retrieval	O
of	O
text	O
from	O
large	O
document	O
collections	O
.Finally	O
TREC1	O
revealed	O
the	O
facts	O
that	O
automatic	O
construction	O
of	O
queries	B-Library
from	O
natural	O
language	O
query	B-Library
statements	O
seems	O
to	O
work	O
.	O
</s>
<s>
Techniques	O
based	O
on	O
natural	B-Language
language	I-Language
processing	I-Language
were	O
no	O
better	O
no	O
worse	O
than	O
those	O
based	O
on	O
vector	O
or	O
probabilistic	O
approach	O
.	O
</s>
<s>
In	O
TREC-6	O
Three	O
new	O
tracks	O
speech	O
,	O
cross	O
language	O
,	O
high	O
precision	O
information	B-Library
retrieval	I-Library
were	O
introduced	O
.	O
</s>
<s>
TREC-7	O
contained	O
seven	O
tracks	O
out	O
of	O
which	O
two	O
were	O
new	O
Query	B-Library
track	O
and	O
very	O
large	O
corpus	O
track	O
.	O
</s>
<s>
TREC-8	O
contain	O
seven	O
tracks	O
out	O
of	O
which	O
two	O
–	O
question	B-Algorithm
answering	I-Algorithm
and	O
web	O
tracks	O
were	O
new	O
.	O
</s>
<s>
Goal	O
:	O
run	O
in	O
parallel	O
CLEF	B-Application
2018	O
,	O
NTCIR-14	O
,	O
TREC	O
2018	O
to	O
develop	O
and	O
tune	O
an	O
IR	B-Library
reproducibility	O
evaluation	O
protocol	O
(	O
new	O
track	O
for	O
2018	O
)	O
.	O
</s>
<s>
Enterprise	B-Application
Track	I-Application
Goal	O
:	O
to	O
study	O
search	O
over	O
the	O
data	O
of	O
an	O
organization	O
to	O
complete	O
some	O
task	O
.	O
</s>
<s>
Entity	B-General_Concept
Track	O
Goal	O
:	O
to	O
perform	O
entity-related	O
search	O
on	O
Web	O
data	O
.	O
</s>
<s>
Cross-Language	B-General_Concept
Track	O
Goal	O
:	O
to	O
investigate	O
the	O
ability	O
of	O
retrieval	O
systems	O
to	O
find	O
documents	O
topically	O
regardless	O
of	O
source	O
language	O
.	O
</s>
<s>
After	O
1999	O
,	O
this	O
track	O
spun	O
off	O
into	O
CLEF	B-Application
.	O
</s>
<s>
FedWeb	B-Operating_System
Track	O
Goal	O
:	O
to	O
select	O
best	O
resources	O
to	O
forward	O
a	O
query	B-Library
to	O
,	O
and	O
merge	O
the	O
results	O
so	O
that	O
most	O
relevant	O
are	O
on	O
the	O
top	O
.	O
</s>
<s>
Federated	O
Web	B-Application
Search	I-Application
Track	O
Goal	O
:	O
to	O
investigate	O
techniques	O
for	O
the	O
selection	O
and	O
combination	O
of	O
search	B-Application
results	I-Application
from	O
a	O
large	O
number	O
of	O
real	O
on-line	O
web	B-Application
search	I-Application
services	O
.	O
</s>
<s>
Interactive	O
Track	O
Goal	O
:	O
to	O
study	O
user	O
interaction	O
with	O
text	B-Algorithm
retrieval	I-Algorithm
systems	O
.	O
</s>
<s>
Legal	O
Track	O
Goal	O
:	O
to	O
develop	O
search	O
technology	O
that	O
meets	O
the	O
needs	O
of	O
lawyers	O
to	O
engage	O
in	O
effective	O
discovery	O
in	O
digital	B-Protocol
document	I-Protocol
collections	O
.	O
</s>
<s>
Microblog	B-Application
Track	O
Goal	O
:	O
to	O
examine	O
the	O
nature	O
of	O
real-time	O
information	O
needs	O
and	O
their	O
satisfaction	O
in	O
the	O
context	O
of	O
microblogging	B-Application
environments	O
such	O
as	O
Twitter	O
.	O
</s>
<s>
Natural	B-Language
language	I-Language
processing	I-Language
Track	O
Goal	O
:	O
to	O
examine	O
how	O
specific	O
tools	O
developed	O
by	O
computational	O
linguists	O
might	O
improve	O
retrieval	O
.	O
</s>
<s>
OpenSearch	O
Track	O
Goal	O
:	O
to	O
explore	O
an	O
evaluation	O
paradigm	O
for	O
IR	B-Library
that	O
involves	O
real	O
users	O
of	O
operational	O
search	B-Application
engines	I-Application
.	O
</s>
<s>
Question	B-Algorithm
Answering	I-Algorithm
Track	O
Goal	O
:	O
to	O
achieve	O
more	O
information	B-Library
retrieval	I-Library
than	O
just	O
document	B-Algorithm
retrieval	I-Algorithm
by	O
answering	O
factoid	O
,	O
list	O
and	O
definition-style	O
questions	O
.	O
</s>
<s>
Relevance	B-Application
Feedback	I-Application
Track	O
Goal	O
:	O
to	O
further	O
deep	O
evaluation	O
of	O
relevance	B-Application
feedback	I-Application
processes	O
.	O
</s>
<s>
Session	O
Track	O
Goal	O
:	O
to	O
develop	O
methods	O
for	O
measuring	O
multiple-query	O
sessions	O
where	O
information	O
needs	O
drift	O
or	O
get	O
more	O
or	O
less	O
specific	O
over	O
the	O
session	O
.	O
</s>
<s>
Goal	O
:	O
to	O
test	O
whether	O
systems	O
can	O
induce	O
the	O
possible	O
tasks	O
users	O
might	O
be	O
trying	O
to	O
accomplish	O
given	O
a	O
query	B-Library
.	O
</s>
<s>
Terabyte	O
Track	O
Goal	O
:	O
to	O
investigate	O
whether/how	O
the	O
IR	B-Library
community	O
can	O
scale	O
traditional	O
IR	B-Library
test-collection-based	O
evaluation	O
to	O
significantly	O
large	O
collections	O
.	O
</s>
<s>
Web	O
Track	O
Goal	O
:	O
to	O
explore	O
information	O
seeking	O
behaviors	O
common	O
in	O
general	O
web	B-Application
search	I-Application
.	O
</s>
<s>
In	O
1997	O
,	O
a	O
Japanese	O
counterpart	O
of	O
TREC	O
was	O
launched	O
(	O
first	O
workshop	O
in	O
1999	O
)	O
,	O
called	O
(	O
NII	O
Test	O
Collection	O
for	O
IR	B-Library
Systems	O
)	O
,	O
and	O
in	O
2000	O
,	O
CLEF	B-Application
,	O
a	O
European	O
counterpart	O
,	O
specifically	O
vectored	O
towards	O
the	O
study	O
of	O
cross-language	B-General_Concept
information	I-General_Concept
retrieval	I-General_Concept
was	O
launched	O
.	O
</s>
<s>
Forum	O
for	O
Information	B-Library
Retrieval	I-Library
Evaluation	O
started	O
in	O
2008	O
with	O
the	O
aim	O
of	O
building	O
a	O
South	O
Asian	O
counterpart	O
for	O
TREC	O
,	O
CLEF	B-Application
,	O
and	O
NTCIR	O
,	O
</s>
<s>
Technology	O
first	O
developed	O
in	O
TREC	O
is	O
now	O
included	O
in	O
many	O
of	O
the	O
world	O
's	O
commercial	O
search	B-Application
engines	I-Application
.	O
</s>
<s>
An	O
independent	O
report	O
by	O
RTII	O
found	O
that	O
"	O
about	O
one-third	O
of	O
the	O
improvement	O
in	O
web	B-Application
search	I-Application
engines	I-Application
from	O
1999	O
to	O
2009	O
is	O
attributable	O
to	O
TREC	O
.	O
</s>
<s>
Those	O
enhancements	O
likely	O
saved	O
up	O
to	O
3	O
billion	O
hours	O
of	O
time	O
using	O
web	B-Application
search	I-Application
engines	I-Application
.	O
</s>
<s>
...	O
Additionally	O
,	O
the	O
report	O
showed	O
that	O
for	O
every	O
$1	O
that	O
NIST	O
and	O
its	O
partners	O
invested	O
in	O
TREC	O
,	O
at	O
least	O
$3.35	O
to	O
$5.07	O
in	O
benefits	O
were	O
accrued	O
to	O
U.S.	O
information	B-Library
retrieval	I-Library
researchers	O
in	O
both	O
the	O
private	O
sector	O
and	O
academia.	O
"	O
</s>
<s>
For	O
example	O
,	O
test	O
collections	O
were	O
created	O
for	O
known-item	O
web	B-Application
search	I-Application
which	O
found	O
improvements	O
from	O
the	O
use	O
of	O
anchor	O
text	O
,	O
title	O
weighting	O
and	O
url	O
length	O
,	O
which	O
were	O
not	O
useful	O
techniques	O
on	O
the	O
older	O
ad	O
hoc	O
test	O
collections	O
.	O
</s>
<s>
In	O
2009	O
,	O
a	O
new	O
billion-page	O
web	O
collection	O
was	O
introduced	O
,	O
and	O
spam	O
filtering	O
was	O
found	O
to	O
be	O
a	O
useful	O
technique	O
for	O
ad	O
hoc	O
web	B-Application
search	I-Application
,	O
unlike	O
in	O
past	O
test	O
collections	O
.	O
</s>
<s>
Hal	O
Varian	O
,	O
Chief	O
Economist	O
at	O
Google	B-Application
,	O
says	O
''	O
Better	O
data	O
makes	O
for	O
better	O
science	O
.	O
</s>
<s>
The	O
history	O
of	O
information	B-Library
retrieval	I-Library
illustrates	O
this	O
principle	O
well	O
,	O
"	O
and	O
describes	O
TREC	O
's	O
contribution	O
.	O
</s>
<s>
The	O
IBM	O
researcher	O
team	O
building	O
IBM	B-Application
Watson	I-Application
(	O
aka	O
DeepQA	B-Application
)	O
,	O
which	O
beat	O
the	O
world	O
's	O
best	O
Jeopardy	O
!	O
</s>
