Towards a benchmark for fact checking with knowledge bases