Uselessness of simple co-occurrence measures for IF&IR — a linguistic point of view.
Dominik Kuropka · 2005
This paper gives a short motivation for the use of similarities of term pairs to improve information filtering and ret rieval methods. Further some simple, but often used co-occurrence b ased measures for the estimation of similarity values of term pai rs are presented. The results of these measures are evaluated against the claims on term similarities, which arise from linguistic aspe cts (inflection, hyponymy, meronymy, composition and word group) between the involved terms. The finding of this paper is the us elessness of simple co-occurrence based measures for the improvement of information filtering und retrieval approaches.