SC-CoMIcs: A Superconductivity Corpus for Materials Informatics

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6753–6760 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6753 SC-CoMIcs: A Superconductivity Corpus for Materials Informatics Kyosuke Yamaguchi1, Ryoji Asahi2, Yutaka Sasaki1 1Toyota Technological Institute 2-12-1 Hisakata, Tenpaku-ku, Nagoya, 468-8511 Japan 2Toyota Central R&D Labs., Inc., 41-1, Yokomichi, Nagakute, Aichi 480-1192, Japan 1{sd19453,yutaka.sasaki}@toyota-ti.ac.jp, 2rasahi@mosk.tytlabs.co.jp Abstract This paper describes a novel corpus tailored for the text mining of superconducting materials in Materials Informatics (MI), named SuperConductivety Corpus for Materials Informatics (SC-CoMIcs). Different from biomedical informatics, there exist very few corpora targeting Materials Science and Engineering (MSE). Especially, there is no sizable corpus which can be used to assist the search of superconducting materials. A team of materials scientists and natural language processing experts jointly designed the annotation and constructed a corpus consisting of manually-annotated 1,000 MSE abstracts related to superconductivity. We conducted experiments on the corpus with a neural Named Entity Recognition (NER) tool. The experimental results show that NER performance over the corpus is around 77% in terms of micro-F1, which is comparable to human annotator agreement rates. Using the trained NER model, we automatically annotated 9,000 abstracts and created a term retrieval tool based on the term similarity. This tool can find superconductivity terms relevant to a query term within a specified Named Entity category, which demonstrates the power of our SC-CoMIcs, efficiently providing knowledge for Materials Informatics applications from rapidly expanding publications. Keywords: Materials Informatics, Superconductivity Corpus, Named Entity Recognition, Materials Text Mining 1. Introduction Recently, Materials Informatics (MI) is a hot topic in the field of Material Science and Engineering (MSE). The rea- son behind this surge of interests is that a lot of materi- als scientists are trying to use machine learning to accel- erate the search of new materials. To find new materials, scientists need to select base materials, adequate doping, and process conditions to synthesize the materials. This is a very costly, time-consuming process. Traditionally, the experience and intuition of researchers often led to unex- pected discovery of new useful materials. MI is studied with a great expectation to change the paradigm in such a long process of materials discovery. There has been, how- ever, a bottleneck, i.e., existence of few datasets for MI to work effectively (Ramprasad et al., 2017). The search of superconducting materials is one of the most challenging issues in MSE. Even after historical dis- covery of high-Tc cuprate superconductors (Bednorz and M¨uller, 1986), materials with a higher transition tempera- ture, higher current flow, and higher processability are re- quired for practical applications (Malozemoff et al., 2005; Foltyn et al., 2007). In materials exploration, experts often consult thousands of published and new articles for getting knowledge to decide a research direction. This can be one of the factors that hinder the efficient search of supercon- ducting materials. In this paper, we propose a new corpus named SuperConductivety Corpus for Materials Informatics (SC- CoMIcs) tailaored for the superconductivity domain, evalu- ate the …

Content truncated to 500 words