A linguistic and statistical approach for extracting knowledge from documents

We present and evaluate an innovating method of automatic indexing. It combines a linguistic analysis of the document to be indexed and a statistical analysis by the singular values decomposition of words in the document. The weighting of words combines advantages of their local and global context as well as their position compared to others terms - the co-occurrence. An application was developed in order to propose assignments topics of documents to a hierarchical referential. Finally, we present experimental results and evaluation carried out on documents of Suez-Environment.