Building a Scientific Concept Hierarchy Database (SCHBase)

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 606–615, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Building a Scientific Concept Hierarchy Database (SCHBASE) Eytan Adar University of Michigan Ann Arbor, MI 48104 eadar@umich.edu Srayan Datta University of Michigan Ann Arbor, MI 48104 srayand@umich.edu Abstract Extracted keyphrases can enhance numer- ous applications ranging from search to tracking the evolution of scientific dis- course. We present SCHBASE, a hier- archical database of keyphrases extracted from large collections of scientific liter- ature. SCHBASE relies on a tendency of scientists to generate new abbrevia- tions that “extend” existing forms as a form of signaling novelty. We demon- strate how these keyphrases/concepts can be extracted, and their viability as a database in relation to existing collections. We further show how keyphrases can be placed into a semantically-meaningful “phylogenetic” structure and describe key features of this structure. The com- plete SCHBASE dataset is available at: http://cond.org/schbase.html. 1 Introduction Due to the immense practical value to Informa- tion Retrieval and other text mining applications, keyphrase extraction has become an extremely popular topic of research. Extracted keyphrases, specifically those derived from scientific literature, support search tasks (Anick, 2003), classification and tagging (Medelyan et al., 2009), informa- tion extraction (Wu and Weld, 2008), and higher- level analysis such as the tracking of influence and dynamics of information propagation (Shi et al., 2010; Ohniwa et al., 2010). In our own work we use the extracted hierarchies to predict scien- tific emergence based on how rapidly new vari- ants emerge. Keyphrases themselves capture a diverse set of scientific language (e.g., methods, techniques, materials, phenomena, processes, dis- eases, devices). Keyphrases, and their uses, have been stud- ied extensively (Gil-Leiva and Alonso-Arroyo, 2007). However, automated keyphrase extrac- tion work has often focused on large-scale statis- tical techniques and ignored the scientific com- munication literature. This literature points to the complex ways in which keyphrases are cre- ated in light of competing demands: expressive- ness, findability, succinct writing, signaling nov- elty, signaling community membership, and so on (Hartley and Kostoff, 2003; Ibrahim, 1989; Grange and Bloom, 2000; Gil-Leiva and Alonso- Arroyo, 2007). Furthermore, the tendency to ex- tract keyphrases through statistical mechanisms often leads to flat keyphrase spaces that make anal- ysis of evolution and emergence difficult. Our contention, and the main motivation be- hind our work, is that we can do better by lever- aging explicit mechanisms adopted by authors in keyphrase generation. Specifically, we focus on a tendency to expand keyphrases by adding terms, coupled with a pressure to abbreviate to retain succinctness. As we argue below, scien- tific communication has evolved the use of ab- breviations to deal with various constraints. Ab- breviations, and acronyms specifically, are rela- tively new in many scientific domains (Grange and Bloom, 2000; Fandrych, 2008) but are now ubiq- uitous (Ibrahim, 1989; Cheng, 2010). Keyphrase selection is often motivated by increasing article findability within a domain (thereby increasing …

Content truncated to 500 words