CHRONOS: News Timeline Summarization

📑 Paper: https://arxiv.org/abs/2501.00888

🌏 Chinese Web Demo: https://modelscope.cn/studios/vickywu1022/CHRONOS

🚀Overview

• We propose CHRONOS, a novel retrieval-based approach to Timeline Summarization (TLS) by iteratively posing questions about the topic and the retrieved documents to generate chronological summaries.
• We construct an up-to-date dataset for open- domain TLS, which surpasses existing public datasets in terms of both size and the duration of timelines.
• Experiments demonstrate that our method is effective on open-domain TLS and achieves comparable results with state-of-the-art methods of closed-domain TLS, with significant improvements in efficiency and scalability.

⚗️ OPEN-TLS Dataset

We release our Open-TLS dataset for open-domain Timeline Summarization.

The target news query is presented in news_keywords.py and the ground truth timeline is presented in data/open/{NEWS_KEYWORD}/timelines.jsonl following the below format:

[["YYY-MM-DDT00:00:00", ["", "", ""]]]

Statistics of Open-TLS are:

🛠 Running CHRONOS

Step 1. Dependencies

pip install -r requirements.txt

Step 2. Exampled Questions Generation

The second step is to construct a topic-questions example pool for datasets in data/ .

python question_exampler.py

Or, you can use our provided data/question_examples.json, which contains examples for the crisis, T17 and Open-TLS datasets.

Step 3. Running CHRONOS

🔥 To be continued...

📝 Citation

@article{wu2025unfoldingheadlineiterativeselfquestioning,
      title={Unfolding the Headline: Iterative Self-Questioning for News Retrieval and Timeline Summarization}, 
      author={Weiqi Wu and Shen Huang and Yong Jiang and Pengjun Xie and Fei Huang and Hai Zhao},
      year={2025},
      eprint={2501.00888},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2501.00888}, 
}