Content-based Map of Science using Cross-lingual Document Embedding - A Comparison of US-Japan Funded Projects

Maps depicting the structure of science help us understand the development of science and technology. However, as it is difficult to apply inter-citation and co-citation analysis to recently published papers and ongoing projects that have few or no references, our previous work developed a content-based map by locating research papers and funding projects using word/document embedding. Because difficulties arise when comparing the content-based map in different languages, this paper improves our content-based map by developing a method for generating multi-dimensional vectors in the same space from cross-lingual (English and Japanese) documents. Using 1,000 IEEE papers, we confirmed a similarity of 0.76 for matching bilingual contents. Finally, we constructed a map from 34,000 projects of the National Science Foundation and Japan Society for the Promotion of Science from 2012 to 2015, and we indicate the findings obtained from a comparison of the US-Japan funded projects.