An exploration of features to improve the generalisability of fake news detection models