Identifying Emergent Research Trends by Key Authors and Phrases

Proceedings of the 27th International Conference on Computational Linguistics, pages 259–269 Santa Fe, New Mexico, USA, August 20-26, 2018. 259 Identifying Emergent Research Trends by Key Authors and Phrases Shenhao Jiang†, Animesh Prasad†, Min-Yen Kan†, Kazunari Sugiyama†‡ †School of Computing, National University of Singapore shenhao jiang@u.nus.edu {animesh|kanmy|sugiyama}@comp.nus.edu.sg ‡National Institute of Informatics, Japan zakugus@nii.ac.jp Abstract Identifying emergent research trends is a key issue for both primary researchers as well as sec- ondary research managers. Such processes can uncover the historical development of an area, and yield insight on developing topics. We propose an embedded trend detection framework for this task which incorporates our bijunctive hypothesis that important phrases are written by important authors within a field and vice versa. By ranking both author and phrase information in a multigraph, our method jointly determines key phrases and authoritative authors. We repre- sent this intermediate output as phrasal embeddings, and feed this to a recurrent neural network (RNN) to compute trend scores that identify research trends. Over two large datasets of scien- tific articles, we demonstrate that our approach successfully detects past trends from the field, outperforming baselines based solely on text centrality or citation. 1 Introduction At no time prior to now has the advance of science – as measured by the rate of growth in scientific publications – bloomed more rapidly. Even practitioners of narrowly-defined, specific fields need help to scan the masses of literature to summarize work and to identify developments likely to spur long-term impact due to the potential problem of information excess (Fujita et al., 2012). The identification of such emergent research trends — “radically novel and relatively fast growing research topic characterized by a certain degree of coherence, and a considerable scientific impact” (Wang, 2017) — is a task that is growing in importance and urgency with the increasingly large scale of scholarly literature. Wang (2017)’s survey reveals two important techniques that characterize the state-of-the-art work on this task. First, text mining — specifically in the guise of topic models — have been employed to yield a summative form of large corpora. Both the Dynamic Topic Model (Blei and Lafferty, 2006) and Author–Topic Model (Rosen-Zvi et al., 2004) cater towards different aspects of the special temporal and authoring aspects of topics in scientific documents, specializing the basic Latent Dirichlet allocation (LDA; Blei et al. (2003)) topic model. Second, dense citations among papers has been used to measure the coherence contributing to a well-defined research trend (Price, 1965). Through appropriate metrics, core works can be identified, from which extracted keywords are treated as a trend’s central concepts. Analyses often establish a co-citation network, as typified by Shibata et al. (2008). These works point to the importance of side information external from the content itself, such as publication date or author identity, and its necessary incorporation into a detection methodology. Our observation is that these attributes often correlate with each other, naturally hinting to a framework where such information can reinforce their signal. We thus take inspiration from other …

Content truncated to 500 words