Augmenting the power of LSI in text retrieval: Singular value rescaling

摘要

This paper presents an analysis of several different LSI (latent semantic indexing) query approaches and proposes a novel rescaling technique, namely singular value rescaling (SVR). Experiments on a standardized TREC data set confirmed the effectiveness of SVR, showing an improvement ratio of 5.9% over the best conventional LSI query approach. In addition, we also compared SVR with another scaling technique in text retrieval called iterative residual rescaling (IRR). Experiments on TREC data set show that SVR performs better than IRR.