Dynamic Topic Mapping Using Latent Semantic Indexing

This paper proposes an approach to provide a dynamic multipoint view of textual documents based on summarization in arbitrary scale in order to produce topic maps. Our approach is based on the latent semantic indexing (LSI) to deal with synonymy and polysemy. Textual resources are decomposed into a set of sentences and then summarized by a set of sentences that are similar to the view of user. A document may have various summaries and by consequence several topic maps according different user interests. The advantage of our method is to be independent to the language used in the source text. Our experimentation shows that the summary text can contains the sentences whose words are different from those used in the user view but their meanings are close to those used in the user point of view