Semantic Movie Recommendations - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 125

to find items matching the individual preferences of users. Recommender systems
address this problem by analyzing a huge amount of data taking into account several
different criteria. Items (potentially unknown yet to the user) are ranked based on
the user's (implicitly and explicitly) defined taste. In order to compute high quality
recommendations comprehensive knowledge is required that covers all the aspects
important for computing the relevance of items.
Recommender system are used in a wide variety of domains, such as online
shops (e.g., fashion), news articles, restaurants, travelling, and entertainment (music,
movies, books). Based on limited (sparse) knowledge about the user recommender
systems suggest items potentially unknown, but helpful to the user. Due to the
complexity of the recommendation task, recommender systems apply sophisticated
machine learning approaches enabling the systems aggregating different types of
data and to extract knowledge useful for computing highly relevant suggestions. In
this chapter we explain semantic recommendation algorithms and show at a concrete
example how the approach can be used for building a personalized semantic movie
recommender system.
The chapter is structured as follows. First, we discuss traditional recommender
approaches and explain the challenges (Sect. 5.2 ). Subsequently, we analyze seman-
tic approaches for managing and processing knowledge. Semantic techniques help us
to overcome the problem of sparse data and allow us the aggregation of comprehen-
sive knowledge collections while computing recommendations. Semantic resources,
datasets as well as mapping and scaling models are needed for representing knowl-
edge in an efficient way. These strategies are discussed in Sect. 5.3 . Then, we explain
how semantic approaches can be applied for building a semantic movie recommender
system (Sect. 5.4 ). In Sect. 5.5 , we present our implemented Semantic Movie Rec-
ommender system and discuss the evaluation results. Finally a conclusion and an
outlook are given.
5.2 Challenges in Recommender Approaches
Traditional recommender approaches are usually classified as collaborative or
content-based [ 1 ]. Collaborative recommenders analyze the user's rating behav-
ior [ 15 ] whereas content-based recommender approaches focus on analyzing the
content-based features of items [ 22 ]. Although these recommender algorithms are
widely used, traditional recommenders show several weaknesses and shortcomings.
The new user problem : In order to provide recommendations meeting the indi-
vidual user preferences the recommender system needs detailed information about
the user. When a new user registers at a recommender system the user must create
a profile describing liked and disliked items. Since most users start with a small
initial profile that does not give complete information about the user's preferences,
the recommendation quality for new users is limited.
The new item problem : Recommender system using Collaborative filtering
algorithm compute the relevance of items based on the ratings of users. Items not