Empirical study of a novel approach to LSI for text categorisation

Latent Semantic Indexing (LSI) is a technique used in Information Retrieval (IR) as an effective tool in correlating and retrieving relevant documents. The authors presented a new philosophy for LSI analysis and evaluation based on the use of image processing tools. In this new approach the Term Document Matrix (TDM) generated in the LSI process is visualized and treated as an image enabling techniques from image processing to be applied. This paper presents a novel extension to this work in which various features of the target databases can be used to predict, and pre-select, search criteria. This latest approach has been evaluated and validated by applying it to a range of sample databases.