Learning « Generalization/Specialization » Relations between Concepts –
Application for Automatically Building Thematic Document Hierarchies
Hermine Njike-Fotzo and Patrick Gallinari
In: RIAO 2004, 26-28 Apr 2004, France.
We introduce a new method for automatically constructing concept hierarchies where the concept nodes follow
a generalization / specialization relation. Starting from a set of concepts automatically extracted from a corpus,
we show how to learn generalization / specialization relations between couples of concepts and how this leads
to the construction of the hierarchy. We present an application of this method for building thematic document
hierarchies similar in spirit to those found on internet portals. We also introduce new criteria for evaluating the
quality of such hierarchies and for comparing them. We finally describe a series of tests performed on
document collections coming from LookSmart and NewScientist hierarchies.