Information Technology Reference
In-Depth Information
Part II
Personalization and Recommendation
Services
Overview
Differing from the use cases that are presented in the
rst part of the topic, where
textual documents are enriched and aggregated to ease users
access, another
approach to address the information overload challenge is to rely on recommen-
dation services. The main principle of recommender systems is to proactively
provide information to users that they might be interested in. For instance, online
retailers such as Amazon.com recommend other products that their customers might
be interested in. Recommender systems hence can inform users about things they
might not be aware of and have not been actively searching for. Two paradigms
dominate the recommender systems
'
'
domain: content-based recommender systems
and collaborative
ltering systems. Content-based recommender systems assume
that systems can successfully discover users
preferences from their liked items
'
contents. They provide suggestions by determining content similarities. Collabo-
rative
'
ltering systems aim to exploit the opinion of people with similar tastes.
Thus, items are recommended when similar users of the recommender system
showed interest in them. Thereby, collaborative
ltering systems are able to rec-
ommend any kind of item disregarding their contents. In addition, recommender
systems may combine both paradigms, obtaining hybrid approaches. In this part of
the topic, we present
ve use cases where different recommendation techniques are
employed.
The
rst recommendation scenario focuses on video recommendation. The
movie industry is a multi-billion dollar business with thousands of new movies
released every year, e.g., by large Hollywood and Bollywood studios, but also by
independent
nding new content
that matches individual preferences is a challenging task. Lommatzsch presents in
Chap. 5 a semantic movie recommender system which takes into account semantic
similarity of movies. He argues that movies are semantically similar when they
share speci
lm makers. Given this large number of movies,
c aspects such as the same directors, actors, or belong to the same
genre. He
rst discusses the challenges in creating such recommender system, then
argues for the exploitation of a graph-based knowledge to provide recommenda-
tions and
nally analyzes the advantages of semantic recommender systems.
Search WWH ::




Custom Search