PERSONA-CHAT

我们提出了 PERSONA-CHAT 数据集，这是一个新的对话数据集，由随机配对的众包工作人员之间的 162,064 个话语组成
并且每个人都要求扮演给定的角色（随机分配，由另一组众包创建）。配对的工人被要求自然地聊天，并在谈话中相互了解。这会产生有趣且引人入胜的对话，我们的代理可以尝试学习模仿。

We present the PERSONA-CHAT dataset, a novel conversational dataset comprising 162,064 utterances exchanged between randomly paired crowdworkers. Each participant was required to adopt a pre-specified persona, which was randomly assigned and developed by a separate group of crowdworkers. The paired workers were instructed to converse naturally and learn about one another throughout the dialogue. This generates intriguing and engaging dialogues that our AI agents can seek to learn and replicate.

搜集汇总

背景与挑战

背景概述

PERSONA-CHAT是一个用于对话生成和多轮问答的文本数据集，包含162,064个话语，由众包工作人员在扮演随机分配角色的情况下自然聊天生成。该数据集旨在模拟真实人际互动，支持个性化对话代理的训练，适用于自然语言处理和AIGC领域的研究与应用。