Evaluating the Factuality of Large Language Models using Large-Scale Knowledge Graphs