Report Abuse

Basic Information

CHRONOS is a research codebase and dataset release for open-domain news timeline summarization. The project proposes a retrieval-based method that iteratively poses questions about a topic and retrieved documents to generate chronological summaries. The repository includes the Open-TLS dataset for long-duration, up-to-date timeline tasks, example question pools, and runnable scripts to reproduce experiments. It provides instructions to install Python requirements, generate example questions with question_exampler.py, and run the main pipeline with main.py to retrieve news, generate timelines, and produce evaluation scores. The README documents required API keys for LLMs and web search and points to an associated paper and a web demo. The code is intended to enable reproducible experiments and evaluation on open-domain timeline summarization tasks.

Links

Categorization

App Details

Features
The repository implements an iterative self-questioning retrieval pipeline for timeline summarization that combines web search and large language models. It ships the Open-TLS dataset with ground-truth timelines and topic query definitions in data/open. Utility scripts include question_exampler.py to build topic-question example pools and main.py to run the end-to-end pipeline with configurable rounds and outputs. The README details where to set API keys in src/model.py, src/searcher.py and src/reader.py for Qwen/GPT, Bing Search, and Jina usage. Outputs are organized as retrieved news, generated timelines, and evaluation scores. The project claims comparable performance to closed-domain TLS with gains in efficiency and scalability and includes a demo reference and a linked arXiv paper.
Use Cases
CHRONOS helps researchers and practitioners reproduce and extend an approach to open-domain timeline summarization. The released Open-TLS dataset supplies larger and longer timelines than prior public sets, enabling evaluation on up-to-date news sequences. The codebase provides a full pipeline to query the web, generate example questions, run iterative retrieval and summarization with configurable rounds, and save retrieved documents and evaluation results. Instructions show how to plug in large language models and web search keys so users can test different LLMs and search settings. The repo is useful for benchmarking, experimenting with iterative retrieval strategies, and building demo apps based on the provided example and modelscope reference.

Please fill the required fields*