MobileCS

我们引入了一种新的 “挑战共享任务”，旨在对半监督和增强的面向任务的对话系统进行基准测试，该系统是为移动运营商的自动化客户服务而构建的。该任务由两个轨道组成: 从对话记录中提取信息 (轨道1) 面向任务的对话系统 (轨道2) 此共享任务的一个重要功能是我们发布大约100K对话框 (中文)，来自中国移动的真实用户和客户服务人员之间的真实对话记录，隐私信息匿名。我们将此数据集称为MobileCS (移动客户服务) dialog数据集，它在大小和性质上都与现有的TOD数据集不同。据我们所知，MobileCS不仅是最大的公开可用的多域TOD数据集，而且还包括现实生活中的人与人之间的数据 (即在现实世界中收集的数据)。为了进行比较，广泛使用的MultiWOZ数据集由10k对话框组成，实际上是模拟数据 (即在绿野仙踪模拟游戏中收集)。We introduce a novel "Challenge Shared Task" aimed at benchmarking semi-supervised and enhanced task-oriented dialogue (TOD) systems built for automated customer service of mobile operators. This task comprises two tracks: Track 1 focuses on information extraction from dialogue records, and Track 2 focuses on task-oriented dialogue systems. A key feature of this shared task is that we release approximately 100K Chinese dialogues, which are real conversation logs between actual users and customer service representatives of China Mobile, with all private information anonymized. We name this dataset the MobileCS (Mobile Customer Service) dialog dataset, which differs from existing TOD datasets in both scale and nature. To the best of our knowledge, MobileCS is not only the largest publicly available multi-domain TOD dataset, but also includes real human-to-human data collected in real-world scenarios. For comparison, the widely used MultiWOZ dataset consists of 10K dialogues, which are actually simulated data gathered via the Wizard-of-Oz simulation game.

搜集汇总

背景与挑战

背景概述

MobileCS是一个面向任务的对话数据集，包含约100K个中文对话，来自中国移动的真实用户与客服之间的真实对话记录，隐私信息已匿名化。该数据集用于半监督和增强的对话系统基准测试，支持两个任务轨道：信息提取和对话系统构建。它是目前最大的公开多域TOD数据集，基于真实世界数据，与模拟数据集如MultiWOZ相比具有显著优势。