Learning from Semantic Flora & Fauna
Gunnar Astrand Grimnes, Alun D. Preece, Peter J. Edwards · 2004
We argue that for a personal agent working within a Semantic Web framework, machine learning is essential. We identify two topologies in the Semantic Web, and refer to these as: (1) semantic forests (disjoint trees) and (2) true semantic webs (complex interconnected graphs). An example of (1) is Citeseer BibTeX mapped to RDF; an example of (2) is FOAF, an RDF representation of people and their relationships. In this paper we explore a number of techniques (na¨ive Bayes, K-NN, ILP, clustering) for learning knowledge that is neither explicitly stated nor deducable from such data. The learned knowledge itself consists of firstclass Semantic Web statements, maximizing its usefulness and re-usability. We also discuss the need for preprocessing and fault tolerance when learning from real distributed Semantic Web data.