Open Visual Knowledge Extraction via Relation-Oriented Multimodality Model Prompting