Reproduce “QMSum: A New Benchmark for Query-based Multi-domain Meeting Summarization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ming Zhong, Da Yin, Tao Yu, Ahmad Zaidi, Mutethia Mutuma, Rahul Jha, Ahmed Hassan Awadallah, Asli Celikyilmaz and 3 more
Researchers from Fudan University, UCLA, Yale University, and Microsoft Research introduce QMSum, a new benchmark dataset for query-based multi-domain meeting summarization, enabling targeted information retrieval from lengthy meeting transcripts. The dataset and an associated two-stage summarization approach provide a foundation for developing more user-centric meeting summarization systems.
Key insights
- The effectiveness of query-based meeting summarization heavily relies on the locator's ability to accurately identify relevant segments of the meeting transcript.
- Large-scale pretrained models like BART demonstrate strong performance in summarization when combined with an effective content locator, indicating the value of pretraining for this task.
- Training models on diverse meeting domains improves their generalization capabilities, but there remains a substantial gap between model and human performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.