Information Technology Reference
In-Depth Information
Aggregation of personal information from several applications [
30
,
40
] and using it
for recommendations has been demonstrated in experimental setups [
4
]. However,
this approach is not easily adoptable as most applications keep their data in 'walled
gardens' where the application provider does not allow to get any user information
out of the system, e.g., no API is offered. Thus, it is not easy to get data for one user
from different applications [
4
,
5
]. In addition, privacy and security issues may occur
and users may not be willing to share passwords to allow the aggregation of data from
different accounts. Other works add meta-knowledge from sources like WordNet to
user profiles to describe similar items, e.g., items from the same domain [
13
,
19
]. Of
course, with the aggregation of user information from different applications a user
could help to build a holistic view of the user, but as the data are hard to get, we
have chosen a more applicable way by using free encyclopedic data as the source for
profile enrichment.
7.6 Discussion and Conclusion
This study shows that with a combination of the semantic tracking system and the
UBO, creation of user interests' profiles becomes simple and effective. With no
visible intervention on the website, detailed tracking of user actions is possible. This
is the main requirement of our tracking system. Of course, beneath the surface the
website structure has to be extended with semantic information using micro-formats
or RDFa. But, relying on the semantic tracking solution, with only a few read articles,
the user profile already reflects general interests of the user and allows us to offer a
personalized news stream filtering the huge amount of articles. While the presented
scenario in Sect.
7.2.1
only showed the tracking of mouse events, the SERUM system
also tracks searches for artists and uses this information for profile creation. As a
search is an explicit action, the artists the user searches for received higher weightage
in the user profile. This complex tracking is unobtrusive and transparent for the user,
which was another requirement of our tracking solution. The management of tracked
information using the UBO allows the usage of this data for future personalization
in different applications. If a user registers for a new application, his previously
collected behavior data can be used to adapt the UI to personal preferences or to
compute recommendations.
We also presented a new semantic recommendation approach using enriched user
profiles with data retrieved from semantic encyclopedic datasets. Our evaluation
shows that depending on the scenario the profile enrichment improves the recommen-
dation quality. Especially in scenarios where the given user profile is very small and
the interests of the user differ from the mean taste of the other users (see Sect.
7.5.4
).
However, evaluation also showed some shortcomings of the presented approach.
Enrichment works very well for users with an unusual taste and in scenarios where
the number of users of an application is low; in these scenarios the enriched pro-
files heighten the recommendation quality. By contrast, enrichment is not helpful for
users with large profiles or a popular music taste. In these cases enrichment blurs the
Search WWH ::
Custom Search