Human-Centered AI in Computational Social Science: Evaluating Automated Annotation with Large Language Models