Fast LSI-based techniques for query expansion in text retrieval systems.

It is widely known that spectral techniques are very effective for document retrieval. Recently, a lot of effort has been spent by researchers to provide a formal mathematical explanation for this effectiveness [3]. Latent Semantic Indexing, in particular, is a text retrieval algorithm based on the spectral analysis of the occurrences of terms in text documents. Despite of its value in improving the quality of a text search, LSI has the drawback of an elevate response time, which makes it unsuitable for on-line search in large collections of documents (e.g., web search engines). In this paper we present two approaches aimed to combine the effectiveness of latent semantic analysis with the efficiency of text matching retrieval, through the technique of query expansion. We show that both approaches have relatively small computational cost and we provide experimental evidence of their ability to improve document retrieval.