The underlying idea is that a FV reflects both the
semantic and linguistic neighbourhoods of a
particular entity. The semantic neighbourhood is
computed based on related entities and direct
properties specified in an ontology, while the
linguistic neighbourhood is based on the colocation
of terms in a domain specific corpus.
Therefore, every entity (classes and individuals) of
the ontologies is associated with a FV to tailor it to
the specific terminology of the text corpus. We
give an overall description of how FVs are
constructed. However, in this paper we do not go
into details of this algorithm (details can be found
in [Tomassen & Strasunskas, 2009]). The extrinsic
quality of FV (i.e., its effect on search
performance) has been investigated in (Strasunskas
& Tomassen, 2008). There we reported an
improvement of the search by more than 10%, on
average. Real users have conducted the
experiment. However, because of variances in the
results (partly explained by the diversity of users),
we needed to assess the intrinsic quality of FV by
evaluating the Feature Vector Construction (FVC)
process. Therefore, in this paper we focus on the