Vector Space Basis Change in Information Retrieval

Abstract

The Vector Space Basis Change (VSBC)is the algebraic operator responsible for change ofbasis and it is parameterized by a transition matrix. Ifwe change the vector space basis, then each vectorcomponent changes depending on this matrix. Thestrategy of VSBC has been shown to be effectivein separating relevant document and irrelevant ones.Recently, using this strategy, some feedback algorithmshave been developed. To build a transition matrix someoptimisation methods have been used. In this paper, wepropose to use a simple, convenient and direct methodto build a transition matrix. Based on this method wedevelop a relevance feedback algorithm. Experimentalresults on a TREC collection show that our proposedmethod is effective and generally superior to knownVSBC-based models. We also show that our proposedmethod gives a statistically significant improvement overthese models.