<s>
Cold	B-General_Concept
start	I-General_Concept
is	O
a	O
potential	O
problem	O
in	O
computer-based	B-General_Concept
information	O
systems	O
which	O
involves	O
a	O
degree	O
of	O
automated	O
data	B-Application
modelling	I-Application
.	O
</s>
<s>
The	O
cold	B-General_Concept
start	I-General_Concept
problem	O
is	O
a	O
well	O
known	O
and	O
well	O
researched	O
problem	O
for	O
recommender	B-Application
systems	I-Application
.	O
</s>
<s>
Recommender	B-Application
systems	I-Application
form	O
a	O
specific	O
type	O
of	O
information	B-Application
filtering	I-Application
(	O
IF	O
)	O
technique	O
that	O
attempts	O
to	O
present	O
information	O
items	O
(	O
e-commerce	O
,	O
films	O
,	O
music	O
,	O
books	O
,	O
news	O
,	O
images	O
,	O
web	O
pages	O
)	O
that	O
are	O
likely	O
of	O
interest	O
to	O
the	O
user	O
.	O
</s>
<s>
Typically	O
,	O
a	O
recommender	B-Application
system	I-Application
compares	O
the	O
user	O
's	O
profile	O
to	O
some	O
reference	O
characteristics	O
.	O
</s>
<s>
These	O
characteristics	O
may	O
be	O
related	O
to	O
item	O
characteristics	O
(	O
content-based	O
filtering	O
)	O
or	O
the	O
user	O
's	O
social	O
environment	O
and	O
past	O
behavior	O
(	O
collaborative	B-Algorithm
filtering	I-Algorithm
)	O
.	O
</s>
<s>
There	O
are	O
three	O
cases	O
of	O
cold	B-General_Concept
start	I-General_Concept
:	O
</s>
<s>
The	O
new	O
community	O
problem	O
,	O
or	O
systemic	O
bootstrapping	O
,	O
refers	O
to	O
the	O
startup	O
of	O
the	O
system	O
,	O
when	O
virtually	O
no	O
information	O
the	O
recommender	B-Application
can	O
rely	O
upon	O
is	O
present	O
.	O
</s>
<s>
The	O
item	O
cold-start	B-General_Concept
problem	O
refers	O
to	O
when	O
items	O
added	O
to	O
the	O
catalogue	O
have	O
either	O
none	O
or	O
very	O
little	O
interactions	O
.	O
</s>
<s>
This	O
constitutes	O
a	O
problem	O
mainly	O
for	O
collaborative	B-Algorithm
filtering	I-Algorithm
algorithms	O
due	O
to	O
the	O
fact	O
that	O
they	O
rely	O
on	O
the	O
item	O
's	O
interactions	O
to	O
make	O
recommendations	O
.	O
</s>
<s>
In	O
the	O
context	O
of	O
cold-start	B-General_Concept
items	O
the	O
popularity	O
bias	O
is	O
important	O
because	O
it	O
might	O
happen	O
that	O
many	O
items	O
,	O
even	O
if	O
they	O
have	O
been	O
in	O
the	O
catalogue	O
for	O
months	O
,	O
received	O
only	O
a	O
few	O
interactions	O
.	O
</s>
<s>
While	O
it	O
is	O
expected	O
that	O
some	O
items	O
will	O
be	O
less	O
popular	O
than	O
others	O
,	O
this	O
issue	O
specifically	O
refers	O
to	O
the	O
fact	O
that	O
the	O
recommender	B-Application
has	O
not	O
enough	O
collaborative	O
information	O
to	O
recommend	O
them	O
in	O
a	O
meaningful	O
and	O
reliable	O
way	O
.	O
</s>
<s>
Since	O
content	O
based	O
recommenders	B-Application
choose	O
which	O
items	O
to	O
recommend	O
based	O
on	O
the	O
feature	O
the	O
items	O
possess	O
,	O
even	O
if	O
no	O
interaction	O
for	O
a	O
new	O
item	O
exist	O
,	O
still	O
its	O
features	O
will	O
allow	O
for	O
a	O
recommendation	O
to	O
be	O
made	O
.	O
</s>
<s>
Content-based	O
algorithms	O
relying	O
on	O
user	O
provided	O
features	O
suffer	O
from	O
the	O
cold-start	B-General_Concept
item	O
problem	O
as	O
well	O
,	O
since	O
for	O
new	O
items	O
if	O
no	O
(	O
or	O
very	O
few	O
)	O
interactions	O
exist	O
,	O
also	O
no	O
(	O
or	O
very	O
few	O
)	O
user	O
reviews	O
and	O
tags	O
will	O
be	O
available	O
.	O
</s>
<s>
The	O
new	O
user	O
case	O
refers	O
to	O
when	O
a	O
new	O
user	O
enrolls	O
in	O
the	O
system	O
and	O
for	O
a	O
certain	O
period	O
of	O
time	O
the	O
recommender	B-Application
has	O
to	O
provide	O
recommendation	O
without	O
relying	O
on	O
the	O
user	O
's	O
past	O
interactions	O
,	O
since	O
none	O
has	O
occurred	O
yet	O
.	O
</s>
<s>
This	O
problem	O
is	O
of	O
particular	O
importance	O
when	O
the	O
recommender	B-Application
is	O
part	O
of	O
the	O
service	O
offered	O
to	O
users	O
,	O
since	O
a	O
user	O
who	O
is	O
faced	O
with	O
recommendations	O
of	O
poor	O
quality	O
might	O
soon	O
decide	O
to	O
stop	O
using	O
the	O
system	O
before	O
providing	O
enough	O
interaction	O
to	O
allow	O
the	O
recommender	B-Application
to	O
understand	O
his/her	O
interests	O
.	O
</s>
<s>
A	O
threshold	O
has	O
to	O
be	O
found	O
between	O
the	O
length	O
of	O
the	O
user	O
registration	O
process	O
,	O
which	O
if	O
too	O
long	O
might	O
indice	O
too	O
many	O
users	O
to	O
abandon	O
it	O
,	O
and	O
the	O
amount	O
of	O
initial	O
data	O
required	O
for	O
the	O
recommender	B-Application
to	O
work	O
properly	O
.	O
</s>
<s>
Similarly	O
to	O
the	O
new	O
items	O
case	O
,	O
not	O
all	O
recommender	B-Application
algorithms	O
are	O
affected	O
in	O
the	O
same	O
way	O
.	O
</s>
<s>
Item-item	B-Application
recommenders	I-Application
will	O
be	O
affected	O
as	O
they	O
rely	O
on	O
user	O
profile	O
to	O
weight	O
how	O
relevant	O
other	O
user	O
's	O
preferences	O
are	O
.	O
</s>
<s>
Collaborative	B-Algorithm
filtering	I-Algorithm
algorithms	O
are	O
the	O
most	O
affected	O
as	O
without	O
interactions	O
no	O
inference	O
can	O
be	O
made	O
about	O
the	O
user	O
's	O
preferences	O
.	O
</s>
<s>
User-user	O
recommender	B-Application
algorithms	O
behave	O
slightly	O
differently	O
.	O
</s>
<s>
Due	O
to	O
the	O
high	O
number	O
of	O
recommender	B-Application
algorithms	O
available	O
as	O
well	O
as	O
system	O
type	O
and	O
characteristics	O
,	O
many	O
strategies	O
to	O
mitigate	O
the	O
cold-start	B-General_Concept
problem	O
have	O
been	O
developed	O
.	O
</s>
<s>
The	O
main	O
approach	O
is	O
to	O
rely	O
on	O
hybrid	O
recommenders	B-Application
,	O
in	O
order	O
to	O
mitigate	O
the	O
disadvantages	O
of	O
one	O
category	O
or	O
model	O
by	O
combining	O
it	O
with	O
another	O
.	O
</s>
<s>
All	O
three	O
categories	O
of	O
cold-start	B-General_Concept
(	O
new	O
community	O
,	O
new	O
item	O
,	O
and	O
new	O
user	O
)	O
have	O
in	O
common	O
the	O
lack	O
of	O
user	O
interactions	O
and	O
presents	O
some	O
commonalities	O
in	O
the	O
strategies	O
available	O
to	O
address	O
them	O
.	O
</s>
<s>
A	O
common	O
strategy	O
when	O
dealing	O
with	O
new	O
items	O
is	O
to	O
couple	O
a	O
collaborative	B-Algorithm
filtering	I-Algorithm
recommender	B-Application
,	O
for	O
warm	O
items	O
,	O
with	O
a	O
content-based	O
filtering	O
recommender	B-Application
,	O
for	O
cold-items	O
.	O
</s>
<s>
While	O
the	O
two	O
algorithms	O
can	O
be	O
combined	O
in	O
different	O
ways	O
,	O
the	O
main	O
drawback	O
of	O
this	O
method	O
is	O
related	O
to	O
the	O
poor	O
recommendation	O
quality	O
often	O
exhibited	O
by	O
content-based	O
recommenders	B-Application
in	O
scenarios	O
where	O
it	O
is	O
difficult	O
to	O
provide	O
a	O
comprehensive	O
description	O
of	O
the	O
item	O
characteristics	O
.	O
</s>
<s>
These	O
techniques	O
are	O
called	O
preference	B-General_Concept
elicitation	I-General_Concept
strategies	O
.	O
</s>
<s>
In	O
both	O
cases	O
,	O
the	O
cold	B-General_Concept
start	I-General_Concept
problem	O
would	O
imply	O
that	O
the	O
user	O
has	O
to	O
dedicate	O
an	O
amount	O
of	O
effort	O
using	O
the	O
system	O
in	O
its	O
'	O
dumb	O
 '	O
state	O
–	O
contributing	O
to	O
the	O
construction	O
of	O
their	O
user	O
profile	O
–	O
before	O
the	O
system	O
can	O
start	O
providing	O
any	O
intelligent	O
recommendations	O
.	O
</s>
<s>
For	O
example	O
MovieLens	O
,	O
a	O
web-based	O
recommender	B-Application
system	I-Application
for	O
movies	O
,	O
asks	O
the	O
user	O
to	O
rate	O
some	O
movies	O
as	O
a	O
part	O
of	O
the	O
registration	O
.	O
</s>
<s>
While	O
preference	B-General_Concept
elicitation	I-General_Concept
strategy	O
are	O
a	O
simple	O
and	O
effective	O
way	O
to	O
deal	O
with	O
new	O
users	O
,	O
the	O
additional	O
requirements	O
during	O
the	O
registration	O
will	O
make	O
the	O
process	O
more	O
time-consuming	O
for	O
the	O
user	O
.	O
</s>
<s>
If	O
,	O
for	O
example	O
,	O
a	O
user	O
has	O
been	O
reading	O
information	O
about	O
a	O
particular	O
music	B-Application
artist	I-Application
from	O
a	O
media	O
portal	O
,	O
then	O
the	O
associated	O
recommender	B-Application
system	I-Application
would	O
automatically	O
propose	O
that	O
artist	O
's	O
releases	O
when	O
the	O
user	O
visits	O
the	O
music	O
store	O
.	O
</s>
<s>
Another	O
of	O
the	O
possible	O
techniques	O
is	O
to	O
apply	O
active	B-General_Concept
learning	I-General_Concept
(	O
machine	O
learning	O
)	O
.	O
</s>
<s>
The	O
main	O
goal	O
of	O
active	B-General_Concept
learning	I-General_Concept
is	O
to	O
guide	O
the	O
user	O
in	O
the	O
preference	B-General_Concept
elicitation	I-General_Concept
process	O
in	O
order	O
to	O
ask	O
him	O
to	O
rate	O
only	O
the	O
items	O
that	O
for	O
the	O
recommender	B-Application
point	O
of	O
view	O
will	O
be	O
the	O
most	O
informative	O
ones	O
.	O
</s>
<s>
The	O
cold	B-General_Concept
start	I-General_Concept
problem	O
is	O
also	O
exhibited	O
by	O
interface	B-Application
agents	B-General_Concept
.	O
</s>
<s>
The	O
cold	B-General_Concept
start	I-General_Concept
problem	O
may	O
be	O
overcome	O
by	O
introducing	O
an	O
element	O
of	O
collaboration	O
amongst	O
agents	B-General_Concept
assisting	O
various	O
users	O
.	O
</s>
<s>
This	O
way	O
,	O
novel	O
situations	O
may	O
be	O
handled	O
by	O
requesting	O
other	O
agents	B-General_Concept
to	O
share	O
what	O
they	O
have	O
already	O
learnt	O
from	O
their	O
respective	O
users	O
.	O
</s>
<s>
Another	O
recent	O
approach	O
which	O
bears	O
similarities	O
with	O
feature	O
mapping	O
is	O
building	O
a	O
hybrid	O
content-based	O
filtering	O
recommender	B-Application
in	O
which	O
features	O
,	O
either	O
of	O
the	O
items	O
or	O
of	O
the	O
users	O
,	O
are	O
weighted	O
according	O
to	O
the	O
user	O
's	O
perception	O
of	O
importance	O
.	O
</s>
<s>
Although	O
various	O
techniques	O
exist	O
to	O
apply	O
feature	O
weighting	O
to	O
user	O
or	O
item	O
features	O
in	O
recommender	B-Application
systems	I-Application
,	O
most	O
of	O
them	O
are	O
from	O
the	O
information	B-Library
retrieval	I-Library
domain	O
like	O
tf	O
–	O
idf	O
,	O
Okapi	O
BM25	O
,	O
only	O
a	O
few	O
have	O
been	O
developed	O
specifically	O
for	O
recommenders	B-Application
.	O
</s>
<s>
Hybrid	O
feature	O
weighting	O
techniques	O
in	O
particular	O
are	O
tailored	O
for	O
the	O
recommender	B-Application
system	I-Application
domain	O
.	O
</s>
<s>
Recently	O
,	O
another	O
approach	O
mitigates	O
the	O
cold	B-General_Concept
start	I-General_Concept
problem	O
by	O
assigning	O
lower	O
constraints	O
to	O
the	O
latent	O
factors	O
associated	O
with	O
the	O
items	O
or	O
users	O
that	O
reveal	O
more	O
information	O
(	O
i.e.	O
,	O
popular	O
items	O
and	O
active	O
users	O
)	O
,	O
and	O
set	O
higher	O
constraints	O
to	O
the	O
others	O
(	O
i.e.	O
,	O
less	O
popular	O
items	O
and	O
inactive	O
users	O
)	O
.	O
</s>
<s>
Differentiating	O
regularization	O
weights	O
can	O
be	O
integrated	O
with	O
the	O
other	O
cold	B-General_Concept
start	I-General_Concept
mitigating	O
strategies	O
.	O
</s>
