Understanding emotions expressed in customer feedback is critical for companies to tailor responses and improve service quality. You have to develop an efficient and explainable NLP system that can accurately interpret the emotional tone of short text messages.

You will work with the Emotion dataset, which contains short text sentences labeled with diverse emotion classes including joy, anger, sadness and more. Your goal is to build models that not only predict these emotions accurately but also provide interpretable explanations regarding how they arrive at their decisions.

To simulate real-world constraints, your solutions must be optimized to run within the resources available, ensuring practicality for deployment in production environments.

You are encouraged to explore approaches that carefully balance predictive performance, model interpretability, and resource efficiency, including methods for handling class imbalance and minimizing inference latency.