Information Technology Reference
In-Depth Information
Fig. 9.7
Matrix factorization. Approximate the user-design matrix
S
by low-rank matrices
X
and
Y
Fig. 9.8
A simplified illustration of the latent factor space generated by matrix factorization. The
latent factors refer to the preferences indicated by the
x
-and
y
-axis. The six users and seven design
elements of Table
9.8
are embedded into the factor space. According to Bartle's player typology
[
2
], users fall into one of the four categories
achiever, explorer, socializer
,and
killer
. Similarly, the
features of the design elements refer to characteristics of the user categories
in order to complete matrix
S
.
To learn the embeddings into the factor space
k
, we need to solve the following
R
basic problem:
s
ug
−
x
u
y
g
2
(
P
)
:
min
x
∗
,
,
y
∗
)
∈
P
(
u
,
g
where
P
is the set of all pairs
(
u
,
g
)
for which
s
ug
is known.
Search WWH ::
Custom Search