NaturalConv
收藏资源简介:
NaturalConv是一个面向多轮主题驱动对话的中文数据集,由腾讯AI实验室创建。该数据集包含约400,000条话语和19,919个对话,平均每轮对话20.1次交流。数据集涵盖体育、娱乐、技术等多个领域,旨在通过自然和流畅的话题转换模拟人类对话。创建过程中,研究人员通过收集新闻文章作为对话基础,允许参与者在保持话题相关性的同时自由扩展话题。NaturalConv适用于评估和提升多轮对话系统的自然性和有效性,特别是在处理复杂和多变的对话场景时。
NaturalConv is a Chinese dataset dedicated to multi-turn topic-driven dialogues, developed by Tencent AI Lab. This dataset contains approximately 400,000 utterances and 19,919 dialogues, with an average of 20.1 exchanges per dialogue session. It covers multiple domains such as sports, entertainment, technology and others, and aims to simulate human-like conversations through natural and fluent topic transitions. During the creation process, researchers collected news articles as the foundation of the dialogues, allowing participants to freely expand topics while maintaining topic relevance. NaturalConv is applicable to evaluating and enhancing the naturalness and effectiveness of multi-turn dialogue systems, particularly in complex and dynamic conversational scenarios.

- 1NaturalConv: A Chinese Dialogue Dataset Towards Multi-turn Topic-driven Conversation腾讯AI实验室 · 2021年



