Weighted similarity: A new similarity measure for document ranking features

Abstract

Many ranking features are utilized by information systems. Several ranking methods act similarly to each other and thus provide similar information. Some information retrieval systems need to select privilege ranking methods and eliminate redundant rankers. To deal with redundant features, the present work introduces a new feature similarity measure, which is based on documents distance. Then the measure is weighted by relevance degree of documents. Experiments are conducted on two data sets MQ2008 and OHSUMED for all features pairs. We adopt two methods of similarity measures in order to compare them with our similarity measure. Results show that our method has correlation with other measures and with MAP.