DTELS: Towards Dynamic Granularity of TimeLine Summarization

📰 News

• Aug. 2025: 🏆 Successfully organized CCKS 2025 Shared Task on Event Timeline Generation for Social Media on Tianchi Competition
• Apr. 2025: 🎉 Paper accepted to NAACL 2025 main conference Proceedings
• Nov. 2024: 📝 Paper preprint published on arXiv:2411.09297

📌 Introduction

We extend the task of Timeline Summarization (TLS) to a new paradigm with timelines at dynamic granularities. We propose a benchmark containing Dataset, Metrics and Evaluations.

📋 Table of Contents

• 🛠️ Installation
• 🚀 Usage
• 🏆 CCKS 2025 Shared Task
• 📁 Project Structure
• 🔧 Troubleshooting
• 📄 License
• ✨ Citation

🛠️ Installation

Prerequisites

• Python 3.10.18

Setup Environment

1. Clone the repository with submodules

git clone --recursive https://github.com/chenlong-clock/DTELS-Bench.git
cd DTELS-Bench

Or if you already cloned without submodules:

git clone https://github.com/chenlong-clock/DTELS-Bench.git
cd DTELS-Bench
git submodule update --init --recursive

2. Create and activate conda environment

conda create -n dtels-env python=3.10
conda activate dtels-env

3. Install required packages

pip install -r requirements.txt

4. Download NLTK data

import nltk
nltk.download('stopwords')

🚀 Usage

Basic Timeline Generation

from utils.data import DTELSArticles
from news_tls.datewise import DatewiseTimelineGenerator, MentionCountDateRanker, PM_Mean_SentenceCollector
from news_tls.summarizers import CentroidOpt

# Load articles data
articles = DTELSArticles(articles_path="articles")

# Initialize timeline generator
generator = DatewiseTimelineGenerator(
    date_ranker=MentionCountDateRanker(),
    sent_collector=PM_Mean_SentenceCollector(),
    summarizer=CentroidOpt()
)

# Generate timeline
timeline = generator.predict(
    collection=articles[1000],  # Use articles from timeline ID 1000
    max_dates=10,
    max_summary_sents=1
)

print("Generated Timeline:")
for item in timeline:
    print(f"Date: {item[1]}, Summary: {item[2]}")

Running the Main Extraction Script

# Extract timelines using different methods
python main_extract.py \
    --method datewise \
    --N 10 \
    --output_path ./extract_output \
    --articles_path ./articles

Command Line Arguments

• --method: Timeline generation method (datewise, clustering, etc.)
• --N: Maximum number of timeline nodes/dates to generate
• --output_path: Directory to save the generated timelines
• --articles_path: Path to the articles directory

🏆 CCKS 2025 Shared Task

This repository includes a simplified evaluation framework for the CCKS 2025 DTELS Shared Task. The evaluation metrics have been streamlined and optimized for competition use while maintaining the core evaluation principles.

⚠️ Important Note on Evaluation Metrics

The ccks2025_demo/ directory contains a simplified version of the DTELS evaluation metrics designed specifically for the CCKS 2025 shared task. This implementation focuses on:

• Simplified Informativeness: Using ROUGE-based matching with Hungarian algorithm
• Streamlined Factuality: Atomic proposition entailment checking
• Basic Granular Consistency: Timeline structure evaluation

For the complete and original DTELS evaluation metrics, please refer to the methodology described in our NAACL 2025 paper. The full implementation follows the detailed guidelines and mathematical formulations presented in the paper.

Quick Start for CCKS 2025 Demo

cd ccks2025_demo/
# Install dependencies (may need to install packages individually)
pip install rank-bm25 rouge-score scipy numpy tqdm

# Run evaluation
bash py_entrance.sh input_param.json eval_result.json

For detailed instructions, see the CCKS 2025 Demo README.

📁 Project Structure

DTELS-Bench/
├── articles/                 # Article data files  
├── ccks2025_demo/           # CCKS 2025 simplified evaluation framework
│   ├── evaluate.py          # Main evaluation orchestrator
│   ├── evaluate_timeline.py # Simplified core evaluation metrics
│   ├── py_entrance.sh       # Entry point script
│   ├── test_data/           # Sample test data
│   └── README.md            # Detailed demo documentation
├── news_tls/                # Core timeline summarization modules
├── utils/                   # Utility functions
├── main_extract.py         # Main extraction script
└── reference_timelines.jsonl # Reference timeline data

🔧 Troubleshooting

time_nlp Submodule Issues

If you encounter issues with the time_nlp module not being visible or importable:

1. Check if submodule is initialized:

git submodule status

2. Initialize/update submodules:

git submodule update --init --recursive

3. If submodule is empty or missing:

git submodule sync
git submodule update --init --recursive

The time_nlp directory is a Git submodule pointing to the Time_NLP project for Chinese time expression recognition.

📄 License

The CCKS 2025 demo evaluation framework is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0). See the LICENSE file for details.

✨Citation

@inproceedings{zhang-etal-2025-dtels,
    title = "{DTELS}: Towards Dynamic Granularity of Timeline Summarization",
    author = "Zhang, Chenlong  and
      Zhou, Tong  and
      Cao, Pengfei  and
      Jin, Zhuoran  and
      Chen, Yubo  and
      Liu, Kang  and
      Zhao, Jun",
    editor = "Chiruzzo, Luis  and
      Ritter, Alan  and
      Wang, Lu",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.naacl-long.136/",
    doi = "10.18653/v1/2025.naacl-long.136",
    pages = "2682--2703",
    ISBN = "979-8-89176-189-6"
}