Semantic Movie Recommendations - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 129

more aspects. In addition to content-based information, IMDb also contains ratings
and reviews. Most of the IMDb's data are available as RDF triples ensuring the
efficient automatic processing of the information.
Discussion : Semantic knowledge resources are valuable sources for recommender
systems because these resources provide different types of knowledge ranging from
fine-grained descriptions of entities (e.g., movies and actors) to ratings and to user
created reviews. Althoughmost resources are not perfectly tailored to a recommender
scenario, themachine readable representation of knowledgemakes it easy to use these
data in a recommender system. The comprehensive collection of data allows semantic
recommender systems to overcome the cold-start problem and to consider a greater
number of aspects than traditional recommender systems. In addition, the graph-
based representation of knowledge and the separation of structure (nodes and edges)
from natural language labels (“node names” might available in different languages)
simplifies the creation of explanations for computed recommendations.
The use of semantic approaches has many advantages; but there are also several
challenges arising from the complexity and the heterogeneity of semantic datasets.
The use of semantically represented data in recommender systems leads to sev-
eral new research questions. Most traditional recommender systems using semantic
datasets focus on datasets having only two node types avoiding problems with the
heterogeneity of edge semantics.
The aggregation of heterogeneous data into one big dataset requires a deep analy-
sis of the different sub-datasets focusing on the meaning of the respective entities
and relationships. In addition, optimized scaling as well as weighting models and
domain-specific edge algebras should be defined reflecting the semantics of the spe-
cific datasets [ 19 ]. Last but not least, the computational complexity of processing
huge graphs must be taken into account when defining a recommender approach for
semantic datasets.
In the next sections we discuss the challenges in detail and present an approach
for learning a universal semantic recommender.
5.3.3 A Dataset for a Semantic Movie Recommender
In order to implement a semantic movie recommender system, we have to find
semantic data sources providing detailed knowledge about movies and all aspects
potentially relevant for computing recommendations. In addition, training data for
optimizing the recommender models are needed. In this section we describe our
semantic movie dataset and discuss the characteristics of the dataset. Subsequently,
we explain step-by-step how to build a powerful semantic recommender based on
the dataset.
The Internet Movie Database (IMDb) is an online database containing informa-
tion related to movies, actors, directors, and production data. IMDb is one of the
most popular online entertainment websites with over 160 million visits each month
[ 12 , 28 ]. Most of the movie data are freely available as Linked Open Data simplifying