In this study, we have evaluated the sensitivity of
the components of a feature vector construction
approach. The overall construction process has
been briefly described analysing its components
w.r.t. both the intrinsic FV quality and three
ontologies used. In total, 32 experiments were conducted. Based on the evaluation of these
experiments we have concluded what components
contribute most positively (the query expansion
and domain identification components) to the FV
quality. The contribution of this paper is a
presentation of an evaluation method and lessons
learnt. We have shown that some choices, when
implementing components, impact the quality of
the resulting FVs and, finally, the performance of
the systems.