KdConv 中文多领域知识驱动的会话数据集

KdConv 全称 Chinese multi-domain Knowledge-driven Conversation dataset，是一个中文多领域知识驱动的对话数据集，将多轮对话中的主题建立在知识图谱上。 KdConv 包含来自三个领域（电影、音乐和旅行）的 4.5 K 个对话，以及平均转数为 19.0 的 86 K 个话语。 该语料库是一个针对多转向知识驱动对话的中国多领域数据集，适用于对多转向人类对话中的知识交互进行建模，包括知识规划，知识基础，知识适应等。KdConv, whose full name is the Chinese multi-domain Knowledge-driven Conversation dataset, is a Chinese multi-domain knowledge-driven dialogue dataset that anchors topics of multi-turn conversations on knowledge graphs. It contains 4.5 thousand dialogues from three domains: film, music, and travel, as well as 86 thousand utterances with an average of 19.0 turns per dialogue. This corpus is a Chinese multi-domain dataset tailored for multi-turn knowledge-driven conversations, suitable for modeling knowledge interactions in multi-turn human conversations, including knowledge planning, knowledge grounding, and knowledge adaptation, among others.

搜集汇总

背景与挑战

背景概述

KdConv是一个中文多领域知识驱动的对话数据集，涵盖电影、音乐和旅行三个领域，包含4.5K个对话和86K个话语，平均转数为19.0，适用于知识驱动对话建模。