A model of opinion dynamics featuring disagreement and biased interactions, which reproduces the phenomenon of echo-chambers, is considered. An improvement of the model presenting open-minded agents, i.e. not acting in a biased way, and their ability to dismantle echo-chambers is also treated. In this setting, we devise a mechanism introducing open-mindedness in a self-consistent way. We also study how open-minded behavior can be enhanced through different mechanisms, leading to the disruption of the echo-chambers. In a second part, we consider a dataset consisting of six months of tweets on italian politics. We build a complete pipeline featuring machine learning and simple natural language processing tools in order to analyse the data.