Multi-XScience: A Large-scale Dataset for Extreme Multi-document Summarization of Scientific Articles

Multi-document summarization is a challenging task for which there exists\nlittle large-scale datasets. We propose Multi-XScience, a large-scale\nmulti-document summarization dataset created from scientific articles.\nMulti-XScience introduces a challenging multi-document summarization task:\nwriting the related-work section of a paper based on its abstract and the\narticles it references. Our work is inspired by extreme summarization, a\ndataset construction protocol that favours abstractive modeling approaches.\nDescriptive statistics and empirical results---using several state-of-the-art\nmodels trained on the Multi-XScience dataset---reveal that Multi-XScience is\nwell suited for abstractive models.\n