The organisation of the ecological transition requires an understanding of the population opinions and what are the acceptable policies, also called "the overton window".
Alhtough it is relatively easy for a human expert to identify the different behaviors and existing opinions, a crucial step is to estimate the strenght of these behaviors and their evolution over time. In this talk, I will describe a simple method to classify twitter accounts on four topics "nuclear power", "meat consumption", "airplane reduction", and "economic degrowht". First, communities are detected, and a few accounts are manually selected to represent the different opinions. Then a non-supervised method based on word-occurence is used to provide a score for the remaining accounts. I will then discuss the perspectives of this work with the integration of human experts.