Clustering for Traceability Managing in System Specifications

System specifications are generally organized according to several documents hierarchies levels linked in order to represent the traceability information. Requirements engineering experts verify manually the links between each specification which allows to generate a traceability matrix. The purpose of this paper is to automatize the generation of the traceability matrix since it is a time consuming and costly task. We propose an artificial intelligence based approach to deal with this problem through a clustering approach. This latter is an unsupervised algorithm that doesn't need any prior knowledge on the language neither the domain of the specifications. Our approach generates duplicates and clusters containing linked requirements. We experiment our approach in an aeronautic domain and a space domain. We obtain better results for high level specifications especially with a pre-processing.