Structural Scaffolds for Citation Intent Classification in Scientific Publications

Proceedings of NAACL-HLT 2019, pages 3586–3596 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 3586 Structural Scaffolds for Citation Intent Classification in Scientific Publications Arman Cohan, Waleed Ammar, Madeleine van Zuylen, Field Cady Allen Institute for Artificial Intelligence {armanc,waleeda,madeleinev,fieldc}@allenai.org Abstract Identifying the intent of a citation in sci- entific papers (e.g., background information, use of methods, comparing results) is criti- cal for machine reading of individual publi- cations and automated analysis of the scien- tific literature. We propose structural scaf- folds, a multitask model to incorporate struc- tural information of scientific papers into ci- tations for effective classification of citation intents. Our model achieves a new state-of- the-art on an existing ACL anthology dataset (ACL-ARC) with a 13.3% absolute increase in F1 score, without relying on external lin- guistic resources or hand-engineered features as done in existing methods. In addition, we introduce a new dataset of citation intents (Sci- Cite) which is more than five times larger and covers multiple scientific domains compared with existing datasets. Our code and data are available at: https://github.com/ allenai/scicite. 1 Introduction Citations play a unique role in scientific discourse and are crucial for understanding and analyzing scientific work (Luukkonen, 1992; Leydesdorff, 1998). They are also typically used as the main measure for assessing impact of scientific pub- lications, venues, and researchers (Li and Ho, 2008). The nature of citations can be different. Some citations indicate direct use of a method while some others merely serve as acknowledg- ing a prior work. Therefore, identifying the in- tent of citations (Figure 1) is critical in improving automated analysis of academic literature and sci- entific impact measurement (Leydesdorff, 1998; Small, 2018). Other applications of citation in- tent classification are enhanced research experi- ence (Moravcsik and Murugesan, 1975), informa- tion retrieval (Ritchie, 2009), summarization (Co- …. A previously described comp- uterized force sensitive system was used to quantify gait cycle timing, specifically the swing time and the stride-to-stride variability of swing time (Bazner et al. 2000). …. Title: Gait asymmetry in patients with Parkinson’s disease and elderly fallers ... Citing paper method background Bazner et al. 2000 Springer et al. 2006 Cited papers … Further details are included in the earlier reports (Springer et al. 2006). …. Figure 1: Example of citations with different intents (BACKGROUND and METHOD). han and Goharian, 2015), and studying evolution of scientific fields (Jurgens et al., 2018). In this work, we approach the problem of ci- tation intent classification by modeling the lan- guage expressed in the citation context. A ci- tation context includes text spans in a citing pa- per describing a referenced work and has been shown to be the primary signal in intent classifi- cation (Teufel et al., 2006; Abu-Jbara et al., 2013; Jurgens et al., 2018). Existing models for this problem are feature-based, modeling the citation context with respect to a set of predefined hand- engineered features (such as linguistic patterns or cue phrases) and ignoring other signals that could improve prediction. In this …

Content truncated to 500 words