In this article we present a study exploiting machine learning methods for sequential struc- tures extraction dedicated to extract semantic relations in call for tender databases on public facilities projects. One of the relationships we consider concerns the impact of a development project. We characterize it as an association between the concepts that define the infrastructure (buildings) and the concepts that define their implantation, namely surfaces. This sequen- tial structure extraction paradigm is considered as a labeling problem of sequential data. A comparative is carried out exploiting several statistical learning techniques. This study demon- strates the robustness of the CRF model for this kind of task when long term characteristics that describe the contexte of occurrence of the labels are taken into account.