The purpose of the work is to define algorithms for the software system of scientific publications analysis, designed to identify research areas and groups of researchers with similar interests within the same university or faculty.
There are many algorithms for solving information extracting problems, but they have some disadvantages regarding the solved problem. Therefore, we developed a proprietary algorithm that consists of four steps: lexical analysis, terminals normalization, entities combining and filtering.
The results of information extracting are used to solve identification problems of authors groups and keywords groups considered as a clustering problem. The analyzed data are presented in the form of graphs of two types: a weighted graph of authors’ interactions and semantic graph of papers. This allows using for the analysis the clustering algorithms based on graph theory and algorithm of stochastic analysis MCL. An analysis of a test articles sample showed that clustering algorithms based on graph theory and algorithm of MCL identified the same clusters, but the algorithm that based on minimum spanning tree was better regarding computational complexity.