F. Yamout, I. Moghrabi (Lebanon), and M. Oakes (UK)
Latent Semantic Indexing, Querying, Relevance FeedbackSingular Value Decomposition.
Latent Semantic Indexing (LSI), which is an extension of the vector model, improves retrieval performance by reducing the dimensions of the term-document space, in an attempt to solve the synonymy, polysemy, and noise problem. Therefore, two documents may be close to each other, even thought they do not share common terms or may be distant from each other, even if they share some common terms. LSI uses a method from linear algebra, the Singular Value Decomposition (SVD) for dimensionality reduction. Querying and Relevance Feedback in LSI perform differently than other models and are expensive in terms of computation. In this paper, simple improvements to the Querying and Relevance Feedback techniques in the LSI model are presented in order to improve their efficiencies without affecting the performance. In addition, a comparison of the LSI query and RF techniques are addressed as well as a detailed description of many of the LSI equations found in the literature.
Important Links:
Go Back
