Databases Reference
In-Depth Information
Preferences change over time
User preferences may also change over time, which falls outside
the model. For example, at eBay, they might be buying a printer,
which makes them only want ink for a short time.
Cost to update
It's also expensive to update the model as you add more data.
The biggest issues are the first two on the list, namely overfitting and
the curse of dimensionality problem. How should you deal with them?
Let's think back to a method you're already familiar with—linear re‐
gression—and build up from there.
Beyond Nearest Neighbor: Machine Learning
Classification
We'll first walk through a simplification of the actual machine learning
algorithm for this—namely we'll build a separate linear regression
model for each item. With each model, we could then predict for a
given user, knowing their attributes, whether they would like the item
corresponding to that model. So one model might be for predicting
whether you like
Mad Men
and another model might be for predicting
whether you would like Bob Dylan.
Denote by
f
i
,
j
user
i
's stated preference for item
j
if you have it (or
user
i
's attribute, if item
j
is a metadata item like age or is_logged_in).
This is a subtle point but can get a bit confusing if you don't internalize
this: you are treating metadata here
also
as if it's an “item.” We men‐
tioned this before, but it's OK if you didn't get it—hopefully it will click
more now. When we said we could predict what you might like, we're
also saying we could use this to predict your
attribute
; i.e., if we didn't
know
if you were a male/female because that was missing data or we
had never asked you, we might be able to predict that.
To let this idea settle even more, assume we have three numeric at‐
tributes for each user, so we have
f
i
,1
,
f
i
,2
,
and
f
i
,3
.
Then to guess user
i
's preference on a new item (we temporarily denote this estimate by
p
i
) we can look for the best choice of
β
k
so that:
p
i
=
β
1
f
1,
i
+
β
2
f
2,
i
+
β
3
f
3,
i
+ϵ
