Research map / Large language models / Language, safety and interpretability
A Pre-Training Based Personalized Dialogue Generation Model with Persona-Sparse Data
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 |
|---|---|---|---|
| Use the proposed persona-conditioned model instead of the strongest baseline | PersonalDialog dialogue generation · PersonalDialog random test set | Acc.: 48.00 -> 32.80 | Hurt |
| Use the proposed persona-conditioned model instead of the strongest baseline | PersonalDialog dialogue generation · PersonalDialog biased test set | Acc.: 61.61 -> 92.13 | Helped |
| Remove pre-training (ablation) | Ours · PersonalDialog random test set | Acc.: 32.80 -> 27.10 | Hurt |
| Remove pre-training (ablation) | Ours · PersonalDialog biased test set | Acc.: 92.13 -> 71.74 | Hurt |
| Add attribute embeddings to the encoder (ablation) | Ours · PersonalDialog random test set | Acc.: 31.85 -> 32.80 | Helped |
| Add attribute embeddings to the encoder (ablation) | Ours · PersonalDialog biased test set | Acc.: 73.51 -> 92.13 | Helped |
| Use a dynamic weight predictor instead of equally averaging attention routes (ablation) | Ours · PersonalDialog random test set | Acc.: 30.70 -> 32.80 | Helped |
| Use a dynamic weight predictor instead of equally averaging attention routes (ablation) | Ours · PersonalDialog biased test set | Acc.: 73.90 -> 92.13 | Helped |
| Replace learned soft weights with heuristic hard weights (ablation) | Ours · PersonalDialog random test set | Acc.: 32.80 -> 32.55 | Hurt |
| Replace learned soft weights with heuristic hard weights (ablation) | Ours · PersonalDialog biased test set | Acc.: 92.13 -> 69.87 | Hurt |
| Set persona weight α to 1 rather than use the predicted weight | Ours · PersonalDialog random test set | Acc.: 32.80 -> 84.55 | Helped |
| Set persona weight α to 0 rather than use the predicted weight | Ours · PersonalDialog random test set | BLEU: 4.18 -> 4.56 | Helped |
Closest papers by meaning
- Personalized Dialogue Generation with Persona-Adaptive Attention (AAAI 2023 · similarity 0.93)
- A Disentangled-Attention Based Framework with Persona-Aware Prompt Learning for Dialogue Generation (AAAI 2023 · similarity 0.93)
- PICLe: Eliciting Diverse Behaviors from Large Language Models with Persona In-Context Learning (ICML 2024 · similarity 0.85)
- PersonaVLM: Long-Term Personalized Multimodal LLMs (CVPR 2026 · similarity 0.84)
- Teaching Language Models to Evolve with Users: Dynamic Profile Modeling for Personalized Alignment (NeurIPS 2025 · similarity 0.84)
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning (NeurIPS 2025 · similarity 0.83)
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.