Towards a Computational Framework for Distinguishing Critical and Conspiratorial Texts by Elaborating on the Context and Argumentation with LLMs.