An Interpretability-Guided Framework for Responsible Synthetic Data Generation in Emotional Text
This work addresses the challenge of data scarcity in emotion recognition for social media analysis, offering a practical solution with critical insights into limitations, though it is incremental in its approach.
The authors tackled the problem of expensive and restricted access to training data for emotion recognition from social media by introducing an interpretability-guided framework using SHAP for LLM-based synthetic data generation, which matched real data performance and improved classification for underrepresented emotion classes but showed reduced vocabulary richness and fewer complex expressions compared to authentic posts.
Emotion recognition from social media is critical for understanding public sentiment, but accessing training data has become prohibitively expensive due to escalating API costs and platform restrictions. We introduce an interpretability-guided framework where Shapley Additive Explanations (SHAP) provide principled guidance for LLM-based synthetic data generation. With sufficient seed data, SHAP-guided approach matches real data performance, significantly outperforms naïve generation, and substantially improves classification for underrepresented emotion classes. However, our linguistic analysis reveals that synthetic text exhibits reduced vocabulary richness and fewer personal or temporally complex expressions than authentic posts. This work provides both a practical framework for responsible synthetic data generation and a critical perspective on its limitations, underscoring that the future of trustworthy AI depends on navigating the trade-offs between synthetic utility and real-world authenticity.