Decision Trees and Recommender Systems - Data Mining with Decision Trees: Theory and Applications - page 256

User's
Gender

Female

Male

Pulp
Fiction

Star
Wars

Dislike

Like

Dislike Like

The Kid
Safety Last!
The Gold Rush
The Big Parade
The General
Metropolis
City Lights
Duck Soup
Se7en
Glory

Paths of Glory
Smultronstället
Vertigo
Touch of Evil
Psycho
The Apartment
The Hustler
Léon
Forrest Gump
Die Hard

Mononoke-hime
Fight Club
The Matrix
American Beauty
Magnolia
Memento
Gladiator
Amores perros
Donnie Darko
Cidade de Deus

The Pianist
Oldboy
Finding Nemo
Big Fish
Mystic River
Forrest Gump
Die Hard
Metropolis
City Lights
Fight Club

Fig. 16.3 An example CF-CB hybrid decision tree.

on-demand recommendations to users. The content based approach requires
n individual trees to be constructed: one for each user. When a user likes to
receive a recommendation, the system needs to traverse the user's tree from
root to leaf once for each item, until it finds an item the user might like
watching. Therefore, the time complexity in this case is O ( h · m ). Similarly,
in the hybrid approach, the tree needs to be traversed once for each item,
andheretoothetimecomplexityis O ( h m ).

In systems that have multiple possible items to recommend and that
require fast computation of recommendations, all the above decision tree
based RS would be impractical. Therefore, it is required to revise the
decision tree to better fit RS, and provide recommendations faster to users.
The proposed algorithm is similar to the ID3 algorithm and uses the hybrid
approach. Because we employ the hybrid approach, only a single tree is
needed, whereas the attributes to split by are user's descriptive attributes.
These qualities can be computed based on the user's past ratings and
the content of the items, as shown in the example in Figure 16.3, but
these can also include user profile attributes aforementioned. The major
variation from the ID3 algorithm is in the tree's leaf nodes; instead of
creating leaf nodes with a label that predicts the target attribute value
(such as rating), each leaf holds a recommendation list. When a user
wishes to receive recommendations, the tree is traversed based on the user's

·