Build a multimodal sentiment analyzer using hint-guided reinforcement learning to combine accuracy with explainability.