Research map / Large language models / Language, safety and interpretability
Rethink Representation Learning for Questionnaire Data
Guanhua Ye, Jifeng He, Yan Li, Junping Du, Zhe Xue, Yingxia Shao, Meiyu Liang, Yawen Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 semantic-wise components (intro and terms, option text, question text) (ablation) | SemantiQ · BRFSS, NLSY, MHP | accuracy change (%): -1.97, -3.73, -3.70 | Improved |
| add feature-wise components (TTT, left-out masking) (ablation) | SemantiQ · BRFSS, NLSY, MHP | accuracy change (%): -1.76, -0.93, -2.54 | Improved |
| add sample-wise ratio-based augmentation (ablation) | SemantiQ · BRFSS, NLSY, MHP | accuracy change (%): -1.60, -1.45, -1.10 | Improved |
| add semantic-wise and feature-wise groups together (ablation) | SemantiQ · BRFSS, NLSY, MHP | accuracy change (%): -3.92, -4.89, -6.55 | Improved |
| add all three component groups (ablation) | SemantiQ · BRFSS, NLSY, MHP | accuracy change (%): -5.60, -6.42, -7.71 | Improved |
| use cell-level SemantiQ instead of row-level TaBERT under column shuffling | TaBERT · BRFSS, NLSY, MHP | accuracy change under shuffled order (%): -6.73 -> -0.98, -5.91 -> -0.31, -7.13 -> -0.79 | Improved |
| use cell-level SemantiQ instead of row-level TabLLM under column shuffling | TabLLM · BRFSS, NLSY, MHP | accuracy change under shuffled order (%): -4.31 -> -0.98, -7.31 -> -0.31, -2.47 -> -0.79 | Improved |
| increase embedding dimension from 16 to 512 (ablation) | SemantiQ · BRFSS | accuracy: about 2.3% improvement | Improved |
| extend LLM statement length beyond 100-150 tokens (ablation) | SemantiQ · BRFSS, NLSY, MHP | accuracy | Hurt |
| apply SemantiQ augmentation to XGBoost (SemantiQ-DT) | XGBoost · BRFSS, NLSY, MHP | accuracy (%): 65.02 -> 69.42, 59.04 -> 63.77, 82.05 -> 90.23 | Improved |
| add semantic upsampling and SemantiQ to FT-Transformer baseline on MHP | FT-Transformer · MHP (5-fold CV) | accuracy (%): 94.01 -> 96.63 | Improved |
| replace row-level LLM calls with cell-level statements | TabLLM/TaBERT · BRFSS | LLM calls: 401,958 -> 107 LLM calls | Improved |
Closest papers by meaning
- Crafting Interpretable Embeddings for Language Neuroscience by Asking LLMs Questions (NeurIPS 2024 · similarity 0.84)
- A General Framework for Producing Interpretable Semantic Text Embeddings (ICLR 2025 · similarity 0.83)
- RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content (NeurIPS 2024 · similarity 0.83)
- A Semantic-Sampling Framework for Evaluating Calibration in Open-Ended Question Answering (NeurIPS 2026 · similarity 0.83)
- NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models (ICLR 2025 · similarity 0.82)
- Ask, Answer, and Detect: Role-Playing LLMs for Personality Detection with Question-Conditioned Mixture-of-Experts (NeurIPS 2026 · similarity 0.82)
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.
