Clusters in Citation Graphs from Research Paper Databases

In citation analysis cluster methods are used to define the directions of researches in some fields or to specify scientific groups. This paper presents the application of two clustering algorithms to citation graphs: the receipt of individual approximated clusters with the usage of the semi-average criterion and structured partitions. Moreover, a comparative analysis with reputable traditional cluster analysis methods is carried out. The key measure of the quality is expert evaluation of the algorithm, based on additional information on the publications (annotations, authors’ names, their place of employment and key words) which was structured in advance and “cleared” by various methods of natural language text processing. The additional goal is to detect the nature of citation with the usage of annotated suffix trees method that contributes to the expert evaluation. The collections of science journals like “Springer. Journal of Classification”, “Elsevier. Journal of Infometrics” and “Springer. Scientometrics” were chosen as data for testing. Besides, a program for exporting information on the publications from the “Socionet.ru” was developed during the work.