Information Technology Reference
In-Depth Information
rated yet by an adequate number of users cannot be recommended by Collaborative
Filtering algorithms. That is the reason why new items (might highly relevant to the
user) are not recommended. To tackle this problem, the recommender system should
apply content-based recommender algorithms, since content-based knowledge does
not depend on user ratings.
The system's cold-start problem
: When setting up a new recommender system,
the number of users and items is limited. Since Collaborative Filtering is based on
the idea of computing the similarity between users and items, the quality of the
recommendations depends on the number of similar-minded users to the current
user. If the number of users in a system is low, the probability is high that no similar
users can be found. This results in reduced recommendation accuracy.
The popularity bias
: Recommender algorithms should assist users in finding
potentially relevant items. Items already known to the user are not good recommen-
dations because they do not mean useful information. Since Collaborative Filtering
algorithm tend to recommend items positively rated by many users, collaborative
filtering algorithms have a strong bias toward popular items. Recommender systems
must be aware of this fact and ensure that at most of the recommendations are new
and useful to the user.
Missing support for multi-lingual data
: Natural language descriptions of con-
tent are a big challenge in the recommendation domain. One the one hand, content-
based description (e.g., movie reviews) are typically available in several different
languages. If a recommender system supports only content in one language, a big
amount of relevant data cannot be processed resulting in sparse data and a low recom-
mendation quality. One the other hand, natural language texts are often ambiguous
requiring detailed linguistic knowledge for resolving ambiguous terms. In order to
overcome the problemof multi-lingual natural language texts, a recommender system
should be able to represent knowledge in a language independent way. In addition,
the system should support the aggregation of knowledge from different languages
ensuring a rich, dense knowledge base. Knowledge extracted from texts in differ-
ent languages as well as from content-based and collaborative knowledge sources
should be represented in a unified data format allowing the efficient management of
heterogeneous knowledge.
One approach for representing knowledge in a universal, natural-language inde-
pendent way is the use of semantic techniques and graph-based approaches. That is
the motivation for us to discuss these approaches in the next section in detail.
5.3 Semantic Approaches and Knowledge Resources
The challenges of missing data, the use of different languages, the need of integrat-
ing different types of knowledge, and the lack of explanations can be solved using
semantic techniques for managing knowledge and for computing recommendations.
The semantic representation of knowledge aims to overcome the problem of tra-
ditionally proprietary data formats tailored to one specific scenario. Semantic data
Search WWH ::
Custom Search