Research map / Large language models / Language, safety and interpretability
Discovering User Attribute Stylistic Differences via Paraphrasing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add paraphrase-pair phrases to the non-paraphrase phrase set (all phrases vs. without paraphrases) (ablation) | Naive Bayes age classifier · Age prediction | ROC AUC: 0.861 -> 0.901 | Helped |
| add paraphrase-pair phrases to the non-paraphrase phrase set (ablation) | Naive Bayes gender classifier · Gender prediction | ROC AUC: 0.784 -> 0.765 | Hurt |
| add paraphrase-pair phrases to the non-paraphrase phrase set (ablation) | Naive Bayes occupational class classifier · Occupational class prediction | ROC AUC: 0.870 -> 0.795 | Hurt |
| use only paraphrase phrases (stylistic proxy) instead of non-paraphrase phrases (topical proxy) (ablation) | Naive Bayes user attribute classifiers · Gender prediction | ROC AUC: 0.784 -> 0.691 | Hurt |
| use only paraphrase phrases instead of non-paraphrase phrases (ablation) | Naive Bayes user attribute classifiers · Occupational class prediction | ROC AUC: 0.870 -> 0.700 | Hurt |
| use only paraphrase phrases instead of non-paraphrase phrases (ablation) | Naive Bayes user attribute classifiers · Age prediction | ROC AUC: 0.861 -> 0.742 | Hurt |
| use paraphrase-based stylistic score against random baseline | user attribute prediction · Gender, age, occupational class | ROC AUC: 0.5 -> 0.691 / 0.742 / 0.700 | Helped |
| apply paraphrase-based style score and compare with human raters | MTurk judgements · Gender, occupational class, age | agreement rate (%): 68.5% / 67.2% / 73.7% | Helped |
Closest papers by meaning
- VibeCheck: Discover and Quantify Qualitative Differences in Large Language Models (ICLR 2025 · similarity 0.84)
- Imitate Before Detect: Aligning Machine Stylistic Preference for Machine-Revised Text Detection (AAAI 2025 · similarity 0.83)
- Uncovering Gaps in How Humans and LLMs Interpret Subjective Language (ICLR 2025 · similarity 0.82)
- Few-Shot Detection of Machine-Generated Text using Style Representations (ICLR 2024 · similarity 0.82)
- Are Models Biased on Text without Gender-related Language? (ICLR 2024 · similarity 0.81)
- ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer (AAAI 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.