Personalized Information Access Using Semantic Knowledge - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 190

us insights into how the application is structured and thus allows us to draw implicit
conclusions from the way the user interacted with it, the interaction gives us explicit
feedback. The click on a recommended item indicates that it matches the users'
interests; to what extent depends on the further interaction. If the user, for instance,
buys an item on a website, this action is a strong indicator for a positive perception,
while a quick return to the recommendation lists indicates that the recommended
item probably did not match the users' interests.
Event :TheOWLclass Event describes the type of event (click, mouse over, etc.) and
the Element or View the user interacted with. An event always occurs on an Element
or View . With the type of the Event , also the time when the event happened is tracked.
This allows to later identify chains of actions and create higher order events. For
example, a click event on an element, followed by a mouse move, followed by a
click release event on a different element could be a Drag-and-Drop event where an
item is dropped into a basket.
7.4 The Enrichment Approach
In this section we present a semantic recommendation approach using the previously
described semantic technologies. We explain the approach in detail and conduct a
comprehensive evaluation to examine how the enrichment influences recommen-
dation quality. Results show that our approach improves recommendation results,
especially for users with uncommon interests.
The general idea of our enrichment approach is visualized in Fig. 7.8 with an
example of a music recommendation system: The figure shows three user profiles
consisting of only a few items without any overlap with the other profiles. In this
case, collaborative filtering (CF) cannot be used. Our profile enrichment process adds
several new items (strongly related to the already present items), so that later, the
user profiles have an overlap and CF can be applied. If a user profile (middle row)
initially contains user interests about 'Björk' and 'Moby,' our enrichment algorithm
takes both entities as input and starts to traverse the semantic dataset which is a graph
where all information is connected. The first entity that is added to the user profile
is the genre entity “electronic,” as both artists are directly connected to it. Then, the
algorithm adds additional artists like “Morcheeba” as the band is also connected to
“electronic.” This enriched user profile is then used for CF.
In this section, we focus on music data to illustrate and evaluate our approach. The
approach itself presented in this section is designed to work on any kind of data as
long as it is presented as a graph. Figure 7.9 shows the general data structure needed
for our approach. The dataset needs a user node that is connected with a like/rated
relation to a set of entities, which can be connected by any kind of relation. The
rate/like relation indicates a positive relation to the linked entity. Negative relations
are currently not considered. The entity nodes can be music information, as in our
scenario, or books, movies, etc.