Generating Event-oriented Attribution for Movies via Two-Stage Prefix-Enhanced Multimodal LLM