---
language:
- en
license:
- cc-by-nc-sa-4.0
multilinguality:
- monolingual
size_categories:
- 100K<n<1M
task_categories:
- summarization
---
# MediaSum
## Description
This large-scale media interview dataset contains 463.6K transcripts with abstractive summaries,
collected from interview transcripts and overview / topic descriptions from NPR and CNN.
### **NOTE: The authors have requested that this dataset be used for research purposes only**
## Homepage
https://github.com/zcgzcgzcg1/MediaSum
## Paper
https://arxiv.org/abs/2103.06410
## Authors
### Chenguang Zhu*, Yang Liu*, Jie Mei, Michael Zeng
#### Microsoft Cognitive Services Research Group
{chezhu,yaliu10,jimei,nzeng}@microsoft.com
## Citation
@article{zhu2021mediasum,
title={MediaSum: A Large-scale Media Interview Dataset for Dialogue Summarization},
author={Zhu, Chenguang and Liu, Yang and Mei, Jie and Zeng, Michael},
journal={arXiv preprint arXiv:2103.06410},
year={2021}
}
## Dataset size
Train: 443,596
Validation: 10,000
Test: 10,000
The splits were made by using the file located here: https://github.com/zcgzcgzcg1/MediaSum/tree/main/data
## Data details
- id (string): unique identifier
- program (string): the program this transcript came from
- date (string): date of program
- url (string): link to where audio and transcript are located
- title (string): title of the program. some datapoints do not have a title
- summary (string): summary of the program
- utt (list of string): list of utterances by the speakers in the program. corresponds with `speaker`
- speaker (list of string): list of speakers, corresponds with `utt`
Example:
```
{
"id": "NPR-11",
"program": "Day to Day",
"date": "2008-06-10",
"url": "https://www.npr.org/templates/story/story.php?storyId=91356794",
"title": "Researchers Find Discriminating Plants",
"summary": "The \"sea rocket\" shows preferential treatment to plants that are its kin. Evolutionary plant ecologist Susan Dudley of McMaster University in Ontario discusses her discovery.",
"utt": [
"This is Day to Day. I'm Madeleine Brand.",
"And I'm Alex Cohen.",
"Coming up, the question of who wrote a famous religious poem turns into a very unchristian battle.",
"First, remember the 1970s? People talked to their houseplants, played them classical music. They were convinced plants were sensuous beings and there was that 1979 movie, \"The Secret Life of Plants.\"",
"Only a few daring individuals, from the scientific establishment, have come forward with offers to replicate his experiments, or test his results. The great majority are content simply to condemn his efforts without taking the trouble to investigate their validity.",
...
"OK. Thank you.",
"That's Susan Dudley. She's an associate professor of biology at McMaster University in Hamilt on Ontario. She discovered that there is a social life of plants."
],
"speaker": [
"MADELEINE BRAND, host",
"ALEX COHEN, host",
"ALEX COHEN, host",
"MADELEINE BRAND, host",
"Unidentified Male",
..."
Professor SUSAN DUDLEY (Biology, McMaster University)",
"MADELEINE BRAND, host"
]
}
```
## Using the dataset
```python
from datasets import load_dataset
ds = load_dataset("nbroad/mediasum")
```
## Data location
https://drive.google.com/file/d/1ZAKZM1cGhEw2A4_n4bGGMYyF8iPjLZni/view?usp=sharing
## License
No license specified, but the authors have requested that this dataset be used for research purposes only.
语言:
- en
许可证:
- cc-by-nc-sa-4.0
多语言属性:
- 单语言
规模类别:
- 100K<n<1M(10万<样本量<100万)
任务类别:
- 摘要生成(summarization)
# MediaSum
## 数据集概述
本大规模媒体访谈数据集包含463.6万条带抽象式摘要(abstractive summaries)的转录文本(transcripts),采集自美国国家公共广播电台(NPR)与美国有线电视新闻网(CNN)的访谈转录文本及概述/主题描述。
### **注意:数据集作者要求本数据集仅可用于科研用途**
## 项目主页
https://github.com/zcgzcgzcg1/MediaSum
## 论文链接
https://arxiv.org/abs/2103.06410
## 作者
### 朱陈光*、刘阳*、梅杰、迈克尔·曾
#### 微软(Microsoft)认知服务研究组
{chezhu,yaliu10,jimei,nzeng}@microsoft.com
## 引用格式
bibtex
@article{zhu2021mediasum,
title={MediaSum: A Large-scale Media Interview Dataset for Dialogue Summarization},
author={Zhu, Chenguang and Liu, Yang and Mei, Jie and Zeng, Michael},
journal={arXiv preprint arXiv:2103.06410},
year={2021}
}
## 数据集规模
训练集:443,596条
验证集:10,000条
测试集:10,000条
数据集划分所使用的配置文件可在此处获取:https://github.com/zcgzcgzcg1/MediaSum/tree/main/data
## 数据字段说明
- id(字符串):唯一标识符
- program(字符串):该转录文本所属的节目名称
- date(字符串):节目播出日期
- url(字符串):音频与转录文本的来源链接
- title(字符串):节目标题,部分数据点无对应标题
- summary(字符串):节目摘要文本
- utt(字符串列表):节目中所有发言者的话语片段(utterances),与`speaker`字段一一对应
- speaker(字符串列表):发言者列表,与`utt`字段一一对应
## 数据示例
json
{
"id": "NPR-11",
"program": "Day to Day",
"date": "2008-06-10",
"url": "https://www.npr.org/templates/story/story.php?storyId=91356794",
"title": "Researchers Find Discriminating Plants",
"summary": ""海棘(sea rocket)"会对亲缘植物表现出偏好性照料。加拿大安大略省麦克马斯特大学的进化植物生态学家苏珊·达德利讲述了她的研究发现。",
"utt": [
"这里是《每日访谈》,我是玛德琳·布兰德。",
"我是亚历克斯·科恩。",
"接下来我们将探讨:一首著名宗教诗歌的作者归属问题如何演变为一场毫不基督的论战。",
"首先,还记得上世纪70年代吗?人们会对着室内盆栽交谈,为它们播放古典音乐。他们坚信植物是拥有感知能力的生命,还有1979年的电影《植物的秘密生活》。",
"仅有少数来自科学界的勇敢人士站出来,表示愿意重复他的实验或验证其研究结果。绝大多数人只是简单地谴责他的研究,却不愿费心去验证其有效性。",
"...",
"好的,谢谢。",
"以上就是苏珊·达德利的分享。她是加拿大安大略省汉密尔顿市麦克马斯特大学生物学副教授,她发现了植物的社交生活。"
],
"speaker": [
"玛德琳·布兰德,主持人",
"亚历克斯·科恩,主持人",
"亚历克斯·科恩,主持人",
"玛德琳·布兰德,主持人",
"匿名男性",
"...",
"苏珊·达德利教授(麦克马斯特大学生物学系)",
"玛德琳·布兰德,主持人"
]
}
## 数据集使用方法
python
from datasets import load_dataset
ds = load_dataset("nbroad/mediasum")
## 数据存储位置
https://drive.google.com/file/d/1ZAKZM1cGhEw2A4_n4bGGMYyF8iPjLZni/view?usp=sharing
## 许可证
未明确指定许可证,但数据集作者要求本数据集仅可用于科研用途。