You are here
An evaluation of new and old similarity ranking algorithms.
The National Library of Medicine's (NLM) IRVIS project has been evaluating “similarity ranking” algorithms that re-order search results according to their similarity to a target result. Several variations of known ranking algorithms were tested, as well as one (we believe) new one which weights terms based on word length. When the algorithms were evaluated using the OHSUMED test collection, the new word length based algorithm was found to outperform the others.