mlsanityData to 14 Oct 2026

An Interpretability-Guided Framework for Responsible Synthetic Data Generation in Emotional Text

arXiv 2511.16132 ↗First posted 20 Nov 2025cs.LGcited by 0

On the map: Synthetic data and self-improvement for LLMs → Privacy-preserving tabular data synthesis

Abstract (arXiv)

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 na"ive 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.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Privacy-preserving tabular data synthesis, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 7 of its 19 references that are in the map. First posted on arXiv 20 Nov 2025.

Builds on

7 of 19 references.

References · 7

  1. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  2. 2015Improving Neural Machine Translation Models with Monolingual Datacited by 788
  3. 2020TweetEval: Unified Benchmark and Comparative Evaluation for Tweet Classificationcited by 417
  4. 2022Generating Training Data with Language Models: Towards Zero-Shot Language Understandingcited by 199
  5. 2022Will we run out of data? Limits of LLM scaling based on human-generated datacited by 195
  6. 2021GPT3Mix: Leveraging Large-scale Language Models for Text Augmentationcited by 187
  7. 2022PromDA: Prompt-based Data Augmentation for Low-Resource NLU Taskscited by 51

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Persona-Based Synthetic Data Generation Using Multi-Stage Conditioning with Large Language Models for Emotion Recognition
  2. 2026Model in Distress: Sentiment Analysis on French Synthetic Social Media
  3. 2025Mechanistic Interpretability of Emotion Inference in Large Language Models
  4. 2026From Emotion Classification to Emotional Reasoning: Enhancing Emotional Intelligence in Large Language Models
  5. 2023Generating Faithful Synthetic Data with Large Language Models: A Case Study in Computational Social Science
  6. 2026NTDH: Complex Reasoning for Comprehensive Affective Analysis
  7. 2024Rethinking Emotion Annotations in the Era of Large Language Models
  8. 2025From Rational Answers to Emotional Resonance: The Role of Controllable Emotion Generation in Language Models