A Systematic Comparison between Extractive Self-Explanations and Human Rationales in Text Classification

1. 1Collect text classification datasets to evaluate self-explanations
2. 2Implement instruction-tuned LLMs to generate self-explanations
3. 3Compare self-explanations with human rationales using plausibility metrics
4. 4Evaluate the quality of self-explanations using human evaluation
5. 5Analyze the results to identify areas for improvement

NLP researchers and AI engineers can benefit from this knowledge to improve the interpretability of their models, while data scientists can use it to evaluate the quality of explanations generated by LLMs

💡 Self-explanations generated by instruction-tuned LLMs can be evaluated using plausibility metrics to determine their quality and effectiveness

🤖 Can LLMs provide good explanations for their outputs? 📊 New research evaluates self-explanations in text classification tasks