Invariance in the structural topic models

The information creation process involving the mass media also uses the source of social networks. The indiscriminate use of this source is very dangerous as it could be exposed to the risk of generalizing the ideas expressed by few in a widespread opinion. However, what people comment on or express on social networks if properly analyzed can help to understand socio-political dynamics. Starting from the collection of tweets concerning the wide reported semantic spectrum of women, we focused on the violence against women and women hate speech (Davidson et al., 2017; Malmasi and Zampieri, 2018). In this paper we propose a invariance study for latent concept, starting from a non-automated keyword selection.