News Timeline Summarization

Data & code for the ACL 2020 paper Examining the State-of-the-Art in News Timeline Summarization (paper, slides).

Updates

Available

• all datasets
• methods & evaluation code
• preprocessing instructions for new datasets

Planned

• instructions to train date ranking models
• more user-friendly fast TLS version to run on unpreprocessed data

Datasets

All datasets used in our experiments are available here, including:

• T17
• Crisis
• Entities

Library installation

The news-tls library contains tools for loading TLS datasets and running TLS methods. To install, run:

pip install -r requirements.txt
pip install -e .

Tilse also needs to be installed for evaluation and some TLS-specific data classes.

Loading a dataset

Check out news_tls/explore_dataset.py to see how to load the provided datasets.

Running methods & evaluation

Check out experiments here.

Format & preprocess your own dataset

If you have a new dataset yourself and want to use preprocess it as the datasets above, check out the preprocessing steps here.

Citation

@inproceedings{gholipour-ghalandari-ifrim-2020-examining,
    title = "Examining the State-of-the-Art in News Timeline Summarization",
    author = "Gholipour Ghalandari, Demian  and
      Ifrim, Georgiana",
    booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.acl-main.122",
    pages = "1322--1334",
}