Title: A development of new metric on graph data using stochastic block model and pointwise mutual information
Authors: Hiroka Hamada - The Institute of Statistical Mathematics (Japan)
Mio Takei - The Institute of Statistical Mathematics (Japan)
Keisuke Honda - The Institute of Statistical Mathematics (Japan) [presenting]
Abstract: A new clustering method is introduced to measure influence of papers in all areas of science and propose a new metric which has well useful properties such ad article level, field independent. To illustrate one application of our method, we analyzed over 7 million articles published between 2012 and 2016 from Web of Science(WOS). Our method consists two key techniques such as stochastic block model (SBM) and Pointwise mutual information (PMI). As first step, to see structure of entire relationship among papers we apply SBM on big scale citation network data. SBM generates a matrix which divides several blocks which represent relationship among research fields. This matrix can be defined as latent research subject based on completely co-citation network of academic activities in real-world. Secondary, to eliminate the influence of bias between research fields we apply PMI as normalization method. Finally, Diversity of papers is calculated by sum of value which correspond to elements in matrix as distance of latent and normalized research filed. The resulting our Research Diversity index (REDi) provides an alternative to the invalid evaluation of using journal impact factors to identify influential papers for various research filed and long-term effect.