This document discusses the "magic barrier" of recommender systems, which refers to the upper bound on recommendation accuracy that is determined by the quality of user rating data. The authors propose that rating coherence, which measures consistency in a user's ratings within specific item features/genres, can serve as a good predictor of the magic barrier. Two experiments are described: 1) Correlating rating coherence with actual magic barrier measurements; 2) Using coherence to improve recommendations by separately training models on coherent vs. incoherent user groups. The results show rating coherence has a strong correlation with magic barrier and using it to separately train recommendation models leads to improved accuracy over treating all users the same.
A Good Girl's Guide to Murder (A Good Girl's Guide to Murder, #1)
The Magic Barrier of Recommender Systems: Rating Coherence Predicts Accuracy
1. The
Magic
Barrier
of
Recommender
Systems:
No
Magic,
Just
ra;ngs
Alejandro
Bellogín,
Alan
Said,
Arjen
de
Vries
@abellogin,
@alansaid,
@arjenpdevries
UMAP
2014
2014-‐07-‐08
3. The
Magic
Barrier
The
upper
bound
on
recommenda;on
accuracy
Recommenda;on
accuracy
is
only
as
good
as
the
quality
of
the
underlying
data
Users
are
noisy
/
incoherent
in
their
ra;ng
behavior
3
4. How
to
find
the
magic
barrier?
1. Collect
infinite
re-‐ra;ngs
for
each
item
from
your
users.
2. Find
standard
devia;on
of
ra;ng/
re-‐ra;ng
inconsistencies
(noise)
[Said
et
al.
UMAP
‘12]
Realis;cally,
an
infinite
amount
of
re-‐ra;ngs
is
not
probable
4
6. What
is
coherence?
Coherent
user
e slightly different. Specifically, the user in Figure 1a gives more consisten
items sharing the same features, which in this case corresponds to the mov
s. In the long term we argue that such a user will have a more consistent (le
ehaviour, since her taste for each item feature seem to be well defined.
Action Comedy
Horror
(a) A coherent user (c(u) = 1.28)
Action Comedy
Horror
(b) A not coherent user (c(u) =
Incoherent
user
htly different. Specifically, the user in Figure 1a gives more consistent rating
sharing the same features, which in this case corresponds to the movie’s gen
he long term we argue that such a user will have a more consistent (less noisy
ur, since her taste for each item feature seem to be well defined.
Action Comedy
Horror
A coherent user (c(u) = 1.28)
Action Comedy
Horror
(b) A not coherent user (c(u) = 5.95)
6
7. DefiniAon
of
Coherence
Ra;ng
consistency
within
a
feature
space
(genres,
themes,
etc.)
c u( )= − σ f
f ∈F
∑ u( )
σ f u( )= r u,i( )− rf u( )( )
2
i∈I u, f( )
∑
Item
features
User’s
ra;ng
devia;on
within
a
feature
7
8. Ra;ng
devia;on
within
a
feature
space
instead
of
infinite
re-‐ra;ngs
Is
this
a
good
es;mate
of
the
magic
barrier?
8
9. Experiments
EX1:
Is
ra;ng
coherence
a
good
predictor
for
the
Magic
Barrier?
EX2:
Can
we
use
ra;ng
coherence
to
improve
recommenda;on
results?
Especially
when
we
do
not
have
re-‐
ra;ngs/magic
barrier?
9
10. RaAng
Coherence
vs.
Magic
Barrier
Rank
users
according
to
magic
barrier/
ra;ng
coherence:
• High/low
magic
barrier
• Low/high
ra;ng
coherence
in
the
“genre”
feature
space
Find
correla;on
(Spearman)
between
magic
barrier/coherence
ranking
*Dataset
from
moviepilot.de,
contains
re-‐ra;ngs
from
306
users
[Said
et
al.
UMAP
‘12]
10
12. Experiments
EX1:
Is
ra;ng
coherence
a
good
predictor
for
the
Magic
Barrier?
EX2:
Can
we
use
ra;ng
coherence
to
improve
recommenda;on
results?
Especially
when
we
do
not
have
re-‐
ra;ngs/magic
barrier?
12
13. Use
raAng
coherence
for
recommendaAon
Divide
users
into
coherent
and
incoherent
–
with
higher
an
lower
magic
barrier
Movielens
1M
Train
each
group
separately
Evaluate
each
group
separately
13
14. RecommendaAon
and
EvaluaAon
Algorithm:
K-‐NN,
Pearson
Training
sets
• “Easy”
users
split
• “Difficult”
users
split
• All
users
split
Evalua;on
sets
• “Easy”
users
split
• “Difficult”
users
split
• All
users
split
Training
Easy
Training
Difficult
Test
Easy
Test
Difficult
14
15. EvaluaAon
Comparisons
All
-‐
All
All
–
Easy
All
–
Difficult
Easy
–
Easy
Difficult
–
Difficult
Training
Easy
Training
Difficult
Test
Easy
Test
Difficult
RMSE:
1.090
User
coverage:
100%
15
16. EvaluaAon
Comparisons
All
-‐
All
All
–
Easy
All
–
Difficult
Easy
–
Easy
Difficult
–
Difficult
Training
Easy
Training
Difficult
Test
Easy
Test
Difficult
RMSE:
0.974
User
coverage:
50%
16
17. EvaluaAon
Comparisons
All
-‐
All
All
–
Easy
All
–
Difficult
Easy
–
Easy
Difficult
–
Difficult
Training
Easy
Training
Difficult
Test
Easy
Test
Difficult
RMSE:
1.195
User
coverage:
50%
17
18. EvaluaAon
Comparisons
All
-‐
All
All
–
Easy
All
–
Difficult
Easy
–
Easy
Difficult
–
Difficult
Training
Easy
Training
Difficult
Test
Easy
Test
Difficult
RMSE:
0.933
User
coverage:
50%
18
19. EvaluaAon
Comparisons
All
-‐
All
All
–
Easy
All
–
Difficult
Easy
–
Easy
Difficult
–
Difficult
Training
Easy
Training
Difficult
Test
Easy
Test
Difficult
RMSE:
1.226
User
coverage:
50%
19
23. Experiments
EX1:
Is
ra;ng
coherence
a
good
predictor
for
the
Magic
Barrier?
EX2:
Can
we
use
ra;ng
coherence
to
improve
recommenda;on
results?
Especially
when
we
do
not
have
re-‐
ra;ngs/magic
barrier?
23
24. Conclusions
Ra;ng
coherence
within
an
item’s
feature
space
is
related
to
the
magic
barrier
Ra;ng
coherence
can
be
used
to
predict
users’
ra;ng
inconsistencies
A
user’s
ra;ng
inconsistencies
tell
us
how
well
our
system
performs
for
that
user
–
i.e.
whether
we
need
to
op;mize
further
24