Fig. 21.8: Cold-start problem in case of social filtering
Table 21.8: Unequal Average Without Misery Strategy with Location unimportant
and threshold 6
We have some evidence that people’s behaviour reflects the outcomes of these
strategies [11], however, more research is clearly needed in this area to see which
strategy is best. Also, more research is needed to establish when to regard a criterion
as ”unimportant”.
The issue of multiple criteria is also the topic of another chapter in this handbook
(see Chapter 24).
21.7.2 Cold-Start Problem
A big problem for recommender systems is the so-called cold-start problem: to adapt
to a user, the system needs to know what the user liked in the past. This is needed
in content-based filtering to decide on items similar to the ones the user liked. It
is needed in social filtering to decide on the users who resemble this user in the
sense that they (dis)liked the same items in the past (see Figure 21.8). So, what if
you do not know anything about the user yet, because they only just started using the
system? Recommender system designers tend to solve this problem by either getting
users to rate items at the start, or by getting them to answer some demographic
questions (and then using stereotypes as a starting point, e.g. elderly people like
classical music).