The paper examines the problem of determining the relevance of documents using knowledge graphs. To solve this problem, an information technology for forming a multilayer knowledge graph reflecting everyday associations of users through weighted associative relations was developed. The implementation of the proposed technology using topic modeling to determine the graph layers is presented. The effectiveness of the proposed implementation was tested using the example of the task of determining the relevance of publications on the VKontakte social network. Experiments have demonstrated the positive effect of using a multilayer knowledge graph in determining relevant publications as part of solving the problem of monitoring the socio-political situation. Possible directions for improving the technology are considered: taking into account metrics of social interest of social media users (likes, views, etc.), studying other methods of identifying graph layers, as well as forming a domain ontology.