Text this: Dynamic Tuning and Multi-Task Learning-Based Model for Multimodal Sentiment Analysis