Research Activity Classification based on Time Series Bibliometrics

TL;DR

This paper develops a machine learning-based method to classify researchers as distinguished or not using features extracted from time series bibliometric data, achieving an 80% F-measure.

Contribution

It introduces a novel approach to classify researchers based on time series bibliometric features, addressing limitations of traditional metrics.

Findings

Achieved 80% F-measure in researcher classification

Demonstrated effectiveness across two research domains

Proposed a new feature extraction method from bibliometric time series

Abstract

Bibliometrics such as the number of papers and times cited are often used to compare researchers based on specific criteria. The criteria, however, are different in each research domain and are set by empirical laws. Moreover, there are arguments, such that the simple sum of metric values works to the advantage of elders. Therefore, this paper attempts to constitute features from time series data of bibliometrics, and then classify the researchers according to the features. In detail, time series patterns are extracted from bibliographic data sets, and then a model to classify whether the researchers are "distinguished" or not is created by a machine learning technique. The experiments achieved an F-measure of 80.0% in the classification of 114 researchers in two research domains based on the data sets of Japan Science and Technology Agency and Elsevier's Scopus. In the future, we will conduct verification on a number of researchers in several domains, and then make use of discovering "distinguished" researchers, who are not widely known.