Алгоритмы выделения групп общения
The purpose of the study: development of methods for analyzing the graph of interacting objects based on the detection of implicit communities in order to solve the problems of searching for the proximity of profiles and the exchange, distribution of information between objects.
Method: importing data from social networks with the subsequent construction of a weighted graph based on the selected attributes and the weight function corresponding to the original task; detection of communities on the constructed weighted graph and comparison of the obtained partitions with the results of classical algorithms.
Results: algorithms to construct graphs and to import attributes were developed and implemented, weight functions created, data structures were constructed, Louvain algorithm for weighted graphs was investigated and implemented with the according to data structures, additional hyper parameters that improve the quality of the standard graph partition by implicit user communities were added. On the example of the social network VKontakte, special algorithms for database crawling are built, the software and hardware complex is applied on real data, and the results of work are compared with the classical algorithms for allocating communities.