Abstract: I will present an automated rumour resolution framework to assist and accelerate the laborious fact-checking process carried by journalists. In this work, I focus on the stance classification, a component of the proposed framework, which could provide some useful insights for the rumour resolution. Thus, I conduct experiments on related documents using three different datasets, namely Fake News Challenge Stage (FNC-1), Football Transfer Rumours 2018 (FTR-18) and Argument Reasoning Comprehension (ARC). My work included cross-dataset and in-domain stance classification experiments, considering the four most frequent topic domains extracted from FNC-1 dataset, celebrities, conflicts, crime and health. The experiments results indicate the difficulty of classifying the minority classes in the analysed datasets and generalising the classification models to different datasets. The in-domain evaluation suggests that some topics (crime) are slightly easier to classify than the others (conflicts).