Improving trust in fact-checking systems with synthetic training data and explanations