What is Helenos?
Helenos is a KDD workbench for the Semantic Web. It tries to find general patterns in unknown data derived from the web, using the ILP learning tool Progol.
Helenos consists of four modules:
- HTML: Fetches html documents from the web and tries to discover general patterns in a dataset of positive and negative examples of documents.
- OWL: Accesses instance data about an ontology and tries to build a rule, which describes the selected individuals.
- XML: The XML module is a proprietary approach to give semantics to xml files by extracting rules from a set of selected nodes.
- XML/OWL: This module allows a mapping between xml elements and ontology individuals in order to enlarge the background knowledge for the rule discovery process.
How does Helenos work?
Helenos transforms the input data into grounded Prolog clauses, which are used as input for the ILP tool Progol.
For example: if a html document has a link to another html document, this is described by the clause "relation(docA,docB)".
Together with a set of modes, Progol approaches the problem of finding general rules by a process called "mode directed inverse entailment". For further information about the implementation and theory of Progol, consult the Progol homepage.
What is the big deal about it?
The task of Inductive Learning is to find hypotheses that are consistent with background knowledge to explain a given set of examples. In general, those hypothesis are definitions of concepts described in some logical language, the examples are descriptions of instances and non-instances of the concept to be learned, and the background knowledge gives additional information about the examples and the concepts' domain knowledge. Having found a pattern in a set of examples, one hopes to be able to extrapolate (or generalize) to unknown examples.
In the context of the Semantic Web, general patterns (or rules) play a crucial part in the "Semantic Web Layer Cake". It would be nice, if the authors of documents explicitly provided us with rules. But usually the documents have to be analyzed in order to find out about the implicitly contained knowledge. This is a tedious and time-consuming process.
The idea of Helenos is to provide a system which finds rules, based on the input of various document types, which can be found in the web.