Research map / Architectures and efficiency / Attention and transformers
Personalized Dialogue Generation with Persona-Adaptive Attention
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 persona-adaptive attention with weighting and dynamic masks | GPT2-SMALL encoder-decoder on ConvAI2 · ConvAI2 | PPL: 17.94 -> 14.03 | Helped |
| use persona-adaptive attention with weighting and dynamic masks | GPT2-SMALL encoder-decoder on ConvAI2 · ConvAI2 | F1: 12.77 -> 17.36 | Helped |
| use persona-adaptive attention with weighting and dynamic masks | persona dialogue models · ConvAI2 | Hits@1: 82.1 -> 93.9 | Helped |
| use persona-adaptive attention instead of GPT2-MEDIUM | GPT2-based dialogue generation · ConvAI2 human evaluation | Fluency: 4.15 -> 4.80 | Helped |
| use persona-adaptive attention instead of GPT2-MEDIUM | GPT2-based dialogue generation · ConvAI2 human evaluation | Informativeness: 3.70 -> 4.54 | Helped |
| use persona-adaptive attention instead of GPT2-MEDIUM | GPT2-based dialogue generation · ConvAI2 human evaluation | Relevance: 3.10 -> 3.69 | Helped |
| use persona-adaptive attention instead of GPT2-MEDIUM | GPT2-based dialogue generation · ConvAI2 human evaluation | Persona consistency: 0.43 -> 0.70 | Helped |
| replace learned balancing with direct sum of the two cross-attentions (ablation) | PAA · ConvAI2 | F1: 17.36 -> 11.37 | Hurt |
| replace learned balancing with a dense layer over concatenated cross-attentions (ablation) | PAA · ConvAI2 | F1: 17.36 -> 12.75 | Hurt |
| use separate weights and masks for persona and context (ablation) | PAA · ConvAI2 | F1: 17.36 -> 15.87 | Hurt |
| add weighting and masking to context cross-attention as well as persona cross-attention (ablation) | PAA · ConvAI2 | F1: 17.30 -> 17.36 | Helped |
| infer weights and masks from persona instead of context cross-attention (ablation) | PAA · ConvAI2 | F1: 17.22 -> 17.36 | Helped |
Closest papers by meaning
- PersonaVLM: Long-Term Personalized Multimodal LLMs (CVPR 2026 · similarity 0.85)
- Teaching Language Models to Evolve with Users: Dynamic Profile Modeling for Personalized Alignment (NeurIPS 2025 · similarity 0.85)
- Enhancing Personalized Multi-Turn Dialogue with Curiosity Reward (NeurIPS 2025 · similarity 0.85)
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
- Mem-PAL: Towards Memory-based Personalized Dialogue Assistants for Long-term User-Agent Interaction (AAAI 2026 · similarity 0.84)
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models (NeurIPS 2026 · 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.