Extraction de termes centrée autour de l'expert
We present a software, named EXIT, built in order to help a field expert to spot relevant terms in texts of his/her specialty. We aim at helping the expert to focus the attention on pertinence. We provide the expert with various statistical measures, and several mining methods. The bottom-up strategy we use to build the terms enables the expert to watch the growth of the terms as new components are included to the shorter terms obtained at iteration n-1. Participating in the TREC'04 Novelty challenge helped us to improve the relevance of the questions asked to the expert.