VI. RECOMMENDATION
From the model defined in Section III, we developed a
recommendation based on object (or video) being consumed.
Two of the most important problems of recommendation
systems area are associated to the recommendation of Best
Item and the Top-N items [14]. The first one consists of
finding, for a specific user, the most interesting item, usually
defined from previously ratings for database items. When
these kind of ratings is not available, and only a list of
purchases or access from an user is known, the problem
turns to find a list or ranking with N potential items of his/her
interest.
Considering our online video environment, we do not have
any type of video rating, but only the user views of each
object. So, the mainly idea of our application is based on
the recommendation of a ranking with potential items (TopN).
We propose to generate this ranking by combining several
object dimensions. According to methodology described in
Section III, such dimensions can be defined from:
• Object: it groups only the object metadata and speci-
fications, as title, description and duration time.
• How/When/Where the object is consumed: it groups
information as popularity, localization and time of
access (consume).
• Who is accessing the object: it groups information
about the user, such as gender and age.
Thus, each item i from our database is compared with all
other items, using a set of these dimensions. The result is a
list of similarities between all the items and i. In the end, our
recommended items are defined after sorting these values to
build a ranking, and chose the N most similar items to i.
These recommendation process can be resumed in:
1) Define the dimensions used for item comparison;
2) Generate a list of similarities between all items from
our database;
3) Sort this list of similarities;