Reproduce “BIGPATENT: A Large-Scale Dataset for Abstractive and Coherent 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
Eva Sharma, Chen Li, Lu Wang
Researchers from Northeastern University and Tencent AI Lab introduce BIGPATENT, a large-scale dataset of 1.3 million patent documents designed to foster truly abstractive and coherent summarization. It mitigates biases found in existing news-based datasets by offering higher abstractiveness, an even distribution of salient content, and a richer discourse structure in its human-written summaries.
Key insights
- BIGPATENT's human-written summaries are significantly more abstractive than those in news datasets, exhibiting higher novel n-gram percentages and lower extractive fragment density.
- Unlike news articles, BIGPATENT displays a remarkably uniform distribution of salient content across the entire input document, making positional baselines ineffective and necessitating global content understanding.
- BIGPATENT summaries feature a richer discourse structure with a higher average number of unique entities and significantly more entity recurrence, demanding models generate more coherent and interwoven narratives.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.