MeetingBank

MeetingBank是一个关于过去十年城市议会会议的新基准数据集。它通过将专业编写的会议记录分成较短的段落，并与会议的特定部分对齐，采用分而治之的方法，将冗长的会议总结分解为更小、更易于管理的任务。该数据集为各种会议总结系统提供了一个新的测试平台，并允许公众了解议会决策的制定过程。MeetingBank is a novel benchmark dataset focused on city council meetings spanning the past decade. Adopting a divide-and-conquer approach, it breaks down lengthy meeting summaries into smaller, more manageable tasks by splitting professionally compiled meeting transcripts into shorter passages and aligning these passages with specific segments of the original meetings. This dataset serves as a new testbed for a diverse range of meeting summarization systems, while allowing the general public to gain insights into the decision-making processes of municipal councils.

搜集汇总

背景与挑战

背景概述

MeetingBank是一个由美国6个主要城市市议会会议构建的基准数据集，包含1366个会议、超过3579小时视频及对应转录文本，并提供了6892个片段级别的摘要实例，适用于会议摘要和视频结构提取研究。