Innovativeness Analysis of Scholarly Publications by Age Prediction Using Ordinal Regression

In this paper we refine our method of measuring the innovativeness of scientific papers. Given a diachronic corpus of papers from a particular field of study, published over a period of a number of years, we extract latent topics and train an ordinal regression model to predict publication years based on topic distributions. Using the prediction error we calculate a real-number based innovation score, which may be used to complement citation analysis in identifying potential breakthrough publications. The innovation score we had proposed previously could not be compared for papers published in different years. The main contribution we make in this work is adjusting the innovation score to account for the publication year, making the scores of papers published in different years directly comparable. We have also improved the prediction accuracy by replacing multiclass classification with ordinal regression and Latent Dirichlet Allocation models with Correlated Topic Models. This also allows for better understanding of the evolution of research topics. We demonstrate our method on two corpora: 3,577 papers published at the International World Wide Web Conference (WWW) between the years 1994 and 2019, and 835 articles published in the Journal of Artificial Societies and Social Simulation (JASSS) from 1998 to 2019.