A Supervised Approach to Extractive Summarisation of Scientific Papers

Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017), pages 195–205, Vancouver, Canada, August 3 - August 4, 2017. c⃝2017 Association for Computational Linguistics A Supervised Approach to Extractive Summarisation of Scientific Papers Ed Collins and Isabelle Augenstein and Sebastian Riedel Department of Computer Science, University College London (UCL), UK {edward.collins.13|i.augenstein|s.riedel}@ucl.ac.uk Abstract Automatic summarisation is a popular ap- proach to reduce a document to its main arguments. Recent research in the area has focused on neural approaches to summari- sation, which can be very data-hungry. However, few large datasets exist and none for the traditionally popular domain of sci- entific publications, which opens up chal- lenging research avenues centered on en- coding large, complex documents. In this paper, we introduce a new dataset for summarisation of computer science pub- lications by exploiting a large resource of author provided summaries and show straightforward ways of extending it fur- ther. We develop models on the dataset making use of both neural sentence en- coding and traditionally used summarisa- tion features and show that models which encode sentences as well as their lo- cal and global context perform best, sig- nificantly outperforming well-established baseline methods. 1 Introduction Automatic summarisation is the task of reducing a document to its main points. There are two streams of summarisation approaches: extractive summarisation, which copies parts of a document (often whole sentences) to form a summary, and abstractive summarisation, which reads a docu- ment and then generates a summary from it, which can contain phrases not appearing in the docu- ment. Abstractive summarisation is the more dif- ficult task, but useful for domains where sentences taken out of context are not a good basis for form- ing a grammatical and coherent summary, like novels. Here, we are concerned with summarising sci- entific publications. Since scientific publications are a technical domain with fairly regular and ex- plicit language, we opt for the task of extractive summarisation. Although there has been work on summarisation of scientific publications before, existing datasets are very small, consisting of tens of documents (Kupiec et al., 1995; Visser and Wieling, 2009). Such small datasets are not suf- ficient to learn supervised summarisation models relying on neural methods for sentence and docu- ment encoding, usually trained on many thousands of documents (Rush et al., 2015; Cheng and Lap- ata, 2016; Chopra et al., 2016; See et al., 2017). In this paper, we introduce a dataset for auto- matic summarisation of computer science publi- cations which can be used for both abstractive and extractive summarisation. It consists of more than 10k documents and can easily be extended auto- matically to an additional 26 domains. The dataset is created by exploiting an existing resource, Sci- enceDirect,1 where many journals require authors to submit highlight statements along with their manuscripts. Using such highlight statements as gold statements has been proven a good gold stan- dard for news documents (Nallapati et al., 2016a). This new dataset offers many exciting research challenges, such how best to encode …

Content truncated to 500 words