Measuring and Reducing LLM Hallucination without Gold-Standard Answers via Expertise-Weighting