Few-Shot Learning for Opinion Summarization