Information Technology Reference
In-Depth Information
Personalized recommendations also help users to discover interesting information
and products based on their preferences and tastes. One challenge is to gather data
about the users' preferences. One way is simply asking the user. This is an obtrusive
way and may lead to losing users who are not willing to put time and effort in training
such an algorithm. Another way is to learn preferences by tracking user interaction.
Finding relevant news on the Internet is becoming increasingly difficult as the
number of news published everyday is exploding. A search on Google News
1
for the
term “Ukraine” returned 61,500 results retrieved in one day. To master this infor-
mation overload, several personalized filtering approaches have been proposed. One
of the first projects was the 1992 started GroupLens project [
16
] that recommended
Usenet news based on collected ratings from other readers. With our web-based appli-
cation SERUM (Semantic Recommendations based on large unstructured datasets),
we support users in finding interesting and up-to-date news about their favorite top-
ics, currently focusing on entertainment news. Therefore, we utilize a broad range
of semantic technologies to further enhance the personalization and recommenda-
tion quality. While other work focus on only one aspect of semantic personalization
support (e.g., [
12
,
39
]), we build a holistic semantic approach, including frontend
and backend solutions, to better learn a user's interest and thus to better recommend
news matching these interests. We incorporate semantic information on the client-
side, using RDFa
2
in the user interface and a user-tracking component that is able
to track this RDFa information [
31
]. In the backend, we have a semantic knowledge
base that includes information from semantic encyclopedic datasets and semantic
technologies that model the users' interest using ontologies to link and enrich them
with semantic information.
In this chapter, we answer the following question: How can semantic tracking and
data management technologies be leveraged for personalization and recommendation
services? In order to address this question, we present SERUM (Semantic Recom-
mendations based on large unstructured datasets), a news recommendation system
that utilizes semantic technologies to collect implicit user behavior and to build
semantic user models. These models, combined with large-scale semantic datasets,
are then used to compute personalized news recommendations using graph-based
algorithms. We introduce the building blocks of SERUM for semantic data man-
agement, personalization, and recommendation, with the main focus on the implicit
user behavior collection. SERUM uses RDFa annotations and a RDFa tracker [
28
]
to collect meaningful user behavior and the User Behavior Ontology (UBO) [
29
],
described in Sect.
7.3
, to build semantic user behavior models. In the following sec-
tions, we first introduce the idea and goal of the SERUM project, followed by an
introduction of the SERUM system. Then, we present the use cases that the semantic
web usage mining approach covers and showcase an example based on the SERUM
system. Finally, we present an evaluation computing on recommendations with a
focus on new users that have not interacted much with the system.
1
http://news.google.com/
,
search conducted on September 19th, 2014.
2
RDFa (or Resource Description Framework—in—attributes) is a W3C Recommendation that
allows to embed rich metadata within Web documents.
Search WWH ::
Custom Search