This paper presents three textual analysis algorithms: the Latent Dirichlet Algorithm, Word2Vec and CamemBERT. The first algorithm groups words from French newspaper articles about the European Central Bank (ECB) in eleven topic categories that can be reduced to two themes: purely economic and political. Word2Vec, helps up to construct a dictionary of positive and negative words. This dictionary is used to compute the tone of the medias. The last one involves Large Langage Models and returns a probability that a news is positive. We use it in a VAR model about monetary policy.