Measuring Researcher Relatedness with Changes in Their Research Interests

Relevant researcher recommendation is important for finding potential research collaborators, and several existing methods measure researcher relatedness based on their research interests. Our previous works represented a researcher with a single multidimensional topic vector calculated from the researcher's publications, ignoring the publication dates. On the other hand, recent studies on information recommendation have shown the effectiveness of modeling changes in user preferences over time. Thus, this paper proposes a new representation of researchers, which consists of yearly topic vectors. To measure the relatedness between researchers, we calculate the similarity between two sequences of topic vectors using Dynamic Time Warping. An experimental example visualizes topic transitions of a target researcher and demonstrates that the proposed method can effectively find researchers whose topic transitions are similar over time, when compared to the conventional method.