Personalized Information Access Using Semantic Knowledge - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 204

7.5.6 Related Work
The Social and Semantic Web has attracted a large number of researchers from
different research fields to find solutions to the cold start problem. So far, different
approaches have been proposed. Approaches range from manipulating the CF process
or manipulating the user model before the CF calculation. In the following section
we present selected works on state-of-the-art CF systems that cope with the cold-
start-problem and present the recent work on user profile enrichment.
7.5.6.1 Collaborative Filtering
In [ 2 ] the authors present an approach that uses existing ontologies, e.g., a movie
ontology, and integrate derived item information with existing user ratings. While
standard CF algorithms assume that all items are distinct, the authors propose an
extended CF algorithm that considers item information as well based on the item
similarity, e.g., the same director. Item similarity is computed by taking into account
similarity between item attributes. To compute the attribute similarities, for each
attribute a similarity function must be defined and an aggregation function that com-
bines the different attribute similarities. This way, it is possible to find similar users
even if they did not rate the same, but similar movies. The approach has the disad-
vantage that it needs effort to build a similarity function for each attribute and it is
also limited to one domain. With our approach we overcome both limitations of this
work. Different weights for different relations/attributes can be learned automati-
cally based on the number of occurrences in the graph, for example, and the domain
limitation is dropped because of our semantic approach where it is easily possible to
bridge different domains.
In a different approach, Middleton et al. [ 23 ] build ontological profiles for users
to recommend research articles. The user profile creation is done using a topic hier-
archy. To overcome the cold-start problem, the authors also attempt to use exter-
nally available information based on personnel records and user publications. The
limitation is that the existence of such additional knowledge cannot be generally
assumed. In some cases, like the presented research community example, public
information is available, but especially on the social web, this information is locked
in the different social networks. Thus, instead of requiring personal information from
external sources, our approach leverages public knowledge sources like Freebase (or
DBpedia).
7.5.6.2 User Profile Enrichment
Different strategies have been proposed to expand the knowledge about users ranging
from the aggregation of user information distributed over different applications to
solutions adding semantic and linguistic knowledge to user profiles [ 14 , 17 , 26 ].