ICML-2005 Tutorial
Machine Learning and the Semantic Web
Andreas Hotho, Steffen Staab
Abstract
Machine Learning and the Semantic Web are two fields of research that
allow for rich and intriguing interactions. Firstly, the Semantic Web
is an effort to put ontologies as well as explicitly and richly
structured semantic data on the WWW for purposes such as query
answering, data integration or intelligent reasoning. While some of its
core building blocks have achieved a quite sophisticated level, e.g.
its representation languages, there is an urgent need to facilitate the
creation of ontologies as well as the creation of corresponding data.
Machine learning here serves as an important vehicle to achieve this
purpose by:
- Wrapper learning
- Semi-automatic semantic annotation
- Ontology learning by classification, clustering and information
extraction
Secondly, such ontologies and semantic data give rise to new
possibilities for improving existing machine learning tasks such as
- Supervised classification
- Unsupervised clustering
Application of such approaches lie in typical applications like text
classification and clustering, but also in improvements and new
possibilities for data integration.
Thirdly, the availability of ontologies and semantic data is not only a
means to improve existing tasks, it also offers chances for defining
new machine learning challenges and applications. The goal of this
tutorial is to acquaint the reader with the basics of the Semantic Web
and then to introduce him to the different interactions between
Semantic Web and Machine Learning that are currently explored. In
addition we will try to address efforts which can be provided by the
machine learning community to bring up the semantic web. There is a
need to support tasks like ontology construction, evolution and mapping
or the filling of the knowledge base. Further resulting semantic data
should to be analyzed by machine learning algorithm. Both direction can
be seen as a challenge for machine learning and can end up in the
development of new algorithm. In the tutorial we will try to address
this issue and we will show possible direction for a fruitful
combination of both areas and future research.
Full Description