News Event Dataset

We provide a collection of news documents labeled at the level of crisp events. How to get the dataset?
The collection spans 4 months (from March 1 to June 30, 2016) and comprises around 147K news documents from 9 popular news channels and published on 3 different online platforms: Twitter, RSS portals, and news websites.
More statistics about the collection are reported in the following table:

events_news_stats

Given a subset of 57 events, we provide relevance labels of news documents for a total of about 4.3K labeled {event, news} pairs. The relevance judgments were collected using a crowdsourcing platform. Examples of events with news documents and their relevance are reported in the follwoing table:

events_news_examples

More details about the collection are available in our short paper published at ICTIR 2019.

Description of the dataset

The dataset consists of 4 files: news_articles.txt, rss_feeds.txt, tweets.txt, and labels.txt

1. News articles

We collected 24,156 news articles. The format of this file is as follows:
document_identifier \t timestamp \t content
where the document_identifier is: {RANDOM_ID}@{PLATFORM}_{CHANNEL} (e.g.,57948@NEWS_abcnews), the timestamp is in the form YYYY-MM-DD hh:mm:ss, and the content was preprocessed.

An extract from the file news_articles.txt

57948@NEWS_abcnews   2016-03-01 00:16:55   ethiopian runners investigated doping sweden ...
57949@NEWS_abcnews   2016-03-01 00:50:31   google says bears responsibility self driving car hit bus australian broadcasting....
57950@NEWS_abcnews   2016-03-01 01:11:22   super tuesday candidates trade insults race white house gets dirty us election 2016...


2. RSS feeds

We collected 43,380 RSS news feeds. The format of this file is as follows:
document_identifier \t timestamp \t content
where the document_identifier is: {RANDOM_ID}@{PLATFORM}_{CHANNEL} (e.g.,613@RSS_abcnews), the timestamp is in the form YYYY-MM-DD hh:mm:ss, and the content was preprocessed.

An extract from the file rss_feeds.txt

613@RSS_abcnews   2016-03-01 00:16:55   ethiopian runners investigated doping
614@RSS_abcnews   2016-03-01 00:50:31   google trialling autonomous cars years prototype vehicle incident lexus first one self driving ...
615@RSS_abcnews   2016-03-01 01:11:22   super tuesday fight gets dirty republican front runner donald trump his closest rival ...


3. Tweets

We collected 80,134 tweets. For copyright reason we do not provide the text of tweets, but we provide their ids. The format of the this file is as follows:
document_identifier \t timestamp
where the document_identifier is: {TWEET_ID}@{PLATFORM}_{CHANNEL} (e.g.,704456822321623040@TWT_abcnews) and the timestamp is in the form YYYY-MM-DD hh:mm:ss.

An extract from the file tweets.txt

704456822321623040@TWT_abcnews   2016-03-01 00:02:49
704460199248031744@TWT_abcnews   2016-03-01 00:16:14
704461948193103872@TWT_abcnews   2016-03-01 00:23:11


4. Labels

For a subset of the news documents, we provide event-relevance labels. Overall, we gathered labels for 4,307 {event, news} pairs using the CrowdFlower crowdsourcing platforms.
The format of the this file is as follows:
document_identifier \t event_keywords \t label (yes or no) \t contributor_confidence (a value from 0.0 to 1.0)

An extract from the file labels.txt

2120@RSS_bbc      japan, kumamoto, earthquake, damage, victims   yes   1.0
4632@RSS_cnn      japan, kumamoto, earthquake, damage, victims   yes   1.0
161876@NEWS_cbc    japan, kumamoto, earthquake, damage, victims   no    0.7193
131333@NEWS_bbc    japan, kumamoto, earthquake, damage, victims   yes   1.0
2128@RSS_bbc      japan, kumamoto, earthquake, damage, victims   yes   1.0



How to get the dataset?

If you're interested in the dataset, please fill in this form and send it to ida DOT mele AT isti DOT cnr DOT it

Please, remember to cite our paper.

Citation

@inproceedings{ICTIR19-MC,
author = {Ida Mele and Fabio Crestani},
title = {A Multi-Source Collection of Event-Labeled News Document},
booktitle = {{The 2019 ACM SIGIR International Conference on the Theory of Information Retrieval}},
series = {{ICTIR'19}},
publisher = {{ACM}},
year = {2019}
}



You can download the BibTeX file here.