PicPersona-TOD
收藏资源简介:
PicPersona-TOD是一个新型面向任务对话数据集,它将用户图像作为个人的一部分,使得系统能够根据用户的特定因素,如年龄或情感背景,提供个性化的响应。该数据集通过用户的第一印象、对话策略指导的提示以及使用外部知识来减少幻觉。人类评估证实,该数据集增强了用户体验,个性化的响应为更吸引人的交互做出了贡献。此外,我们还介绍了一个新的NLG模型Pictor,它不仅能够个性化响应,而且在未见过的领域也表现出稳健的性能。
PicPersona-TOD is a novel task-oriented dialogue dataset that incorporates user images as part of user personas, enabling the system to generate personalized responses based on user-specific attributes such as age or emotional background. This dataset reduces hallucinations by leveraging first impressions of users, prompts guided by dialogue strategies, and external knowledge. Human evaluations confirm that this dataset enhances user experience, and personalized responses contribute to more engaging interactions. Additionally, we introduce a new NLG model named Pictor, which not only generates personalized responses but also exhibits robust performance in unseen domains.
PicPersona-TOD数据集概述
基本信息
- 数据集名称: PicPersona-TOD
- 用途: 用于在任务导向对话中基于图像人物信息生成个性化回复
- 发布状态: 已被NAACL 2025接受
- 作者: Jihyun Lee, Yejin Jeon, Seungyeon Seo, Gary Geunbae Lee
数据集内容
- 数据位置:
/dataset/mwoz_sgd - 数据特点: 包含用于个性化回复生成的图像人物信息
引用信息
- 引用文献: 待定(TBD)
联系方式
- 联系人邮箱: jihyunlee@postech.ac.kr

- 1PicPersona-TOD : A Dataset for Personalizing Utterance Style in Task-Oriented Dialogue with Image PersonaPOSTECH(浦项科技大学) · 2025年



