Decision Trees and Recommender Systems - Data Mining with Decision Trees: Theory and Applications - page 260

profiles as part of the sign up process by asking them to provide feedback
regarding several items. In this section, we present an anytime preferences
elicitation method that uses the idea of pairwise comparison between items.
Our method uses a lazy decision tree, with pairwise comparisons at the
decision nodes. Based on the user's response to a certain co mparison we
select on-the-fly what pairwise comparison should next be asked. A com-
parative field study shows that the proposed pairwise approach provides
more accurate recommendations than existing methods and requires less
effort when signing up newcomers [ Rokach and Kisilevich (2012) ] .
Most recommendation techniques use some types of a user profile or
user model but these techniques cannot provide personalized recommenda-
tions until a user profile has been created. As the user interacts with the
system, the profile can be incrementally established to eventually provide
personalized recommendations. One way to address the challenge is to
ask newcomers to fill-in a simple questionnaire that will lead them to an
immediately beneficial recommendation.
Methods that ask the user to specify her preferences for item attributes
are also known as preference elicitation or preference-based search. Research
in the area has mainly focused on using a set of examples (e.g. the number
of movies the user likes) or through a form of specifying the user's interests.
Such approaches have drawbacks. While rating observed items is a painless
process, using only a set of rated items can cause the system to later
recommend only items similar to the ones the user rated. Furthermore,
asking users to fill in lengthy forms is usually considered a boring chore
and users tend to either avoid it or answer questions arbitrarily (e.g. always
picking the first answer).
Decision trees can improve the initialization of the profile generation
by using a questionnaire. The questionnaire is created as an interactive,
easy-to-use process. At each stage the user is presented with two items and
is asked to select the preferred item. The items are presented as pictures to
make the answering process intuitive. The underlying process is “anytime”
in the sense that although the user may choose to abandon the questionnaire
at any stage, the system is still able to create a profile. The more answers
the user provides, the more specific her profile becomes. It is advisable to
perform a pairwise comparison because previous research shows that users
are more accurate when making relative indirect judgments than when they
directly rank items using a linear scale.
Pairwise comparison is a well-known method in decision making. But
there are only a few attempts to use it in RSs and none of them has been
used in the context of the collaborative filtering setting.