Multilingual author matching across different academic databases: a case study on KAKEN, DBLP, and PubMed

Abstract

Researchers often use their native languages to present and exchange ideas. To construct an individual author’s complete profile, a list of their English and non-English academic publications must be constructed. This paper presents a practical approach for multilingual author matching across different academic databases. Our approach automatically links the academic records of a target database to a researcher identifier of a source database. First, we extracted a comprehensive set of records in the target database, whose author names were identical to the researcher names in the source database. Then, we calculated multiple author similarity measures, which can be adopted in certain entity pairs from different language databases. Finally, we aggregated the measures to output an improved score that indicates the likelihood of each record as being the researcher’s work. Our method was found to be easy to implement, and its performance was evaluated in real database management settings. Experiments were conducted using DBLP and PubMed as the target English databases. As the Japanese database, KAKEN was the source for identifying researcher information. The results demonstrated each similarity measure’s performance, from which we observed that the score aggregation achieved stable performance. Our method can lessen human efforts to associate various scholarly contributions.