Benchmarking Large Language Models in Complex Question Answering Attribution using Knowledge Graphs