Decision Trees and Recommender Systems - Data Mining with Decision Trees: Theory and Applications - page 252

to complete such a computational task, RSs collect from users their
preferences, which are either explicitly expressed, e.g. as products ratings,
or are inferred by interpreting the user's actions. For instance, an RS may
consider the navigation to a particular product page as an implicit sign of
preference for the items shown on that page.
RSs' development has its origins in a rather simple observation:
individuals often rely on recommendations provided by others in making
mundane decisions. For example, it is common to rely on what one's
peers recommend when selecting a topic to read; employers count on
recommendation letters in their recruiting decisions; and when selecting
a movie to watch, individuals tend to trust the movie reviews that a film
critic has written and which appear in the newspaper they read.
In seeking to mimic this behavior, the first RSs applied algorithms to
leverage recommendations produced by a community of users to deliver
recommendations to an active user, i.e. a user looking for suggestions.
The recommendations were for items that similar users (those with similar
tastes) had liked. This approach is termed collaborative-filtering (CF) and
its rationale is that if the active user agreed in the past with some users,
then the other recommendations coming from these similar users should be
relevant as well and of interest to the active user.
16.2 Using Decision Trees for Recommending Items
Decision Trees are used as a model-based approach for recommender sys-
tems. The use of decision trees for building recommendation models offers
several benefits, such as: eciency, interpretability and flexibility in han-
dling a variety of input data types (ratings, demographic, contextual, etc.).
However, simply using existing decision tree induction algorithms for
recommendation tasks is not always straightforward. A major weakness in
using decision trees as a prediction model in RS is the need to build a
huge number of trees (either for each item or for each user). Moreover, the
model can only compute the expected rating of a single item at a time. To
provide recommendations to the user, we must traverse the tree(s) from
root to leaf once for each item in order to compute its predicted rating.
Only after computing the predicted rating of all items can the RS provide
the recommendations (highest predicted rating items). Thus, decision trees
in RS do not scale well with respect to the number of items.
Gershman et al . (2010) propose a modification to the decision tree
model, to make it of practical use for larger scale RS. Instead of predicting