Analysis of Socio-political Content of Internet Resources Based on Hybrid Neural Network Technologies

Abstract

Countering sociocultural and cyber threats, as well as ideological extremism, requires identifying undesirable content of information sources, primarily the content of Internet resources. Significant volumes, as well as semantic and lexicological diversity of Internet content, make it necessary to improve artificial intelligence (AI) methods for neural network analysis in order to identify potential threats and undesirable content. The problem is complicated by the presence of “information garbage” in the studied content, which is a specific information noise. It can be solved using neural network technologies, including frequency preprocessing of text arrays, justification of the structure and construction of a subject-oriented database of text data bodies, justification and experimental study of the architecture of the hybrid ANN based on the vector representation of pre-formed dictionaries of terms. The authors substantiate a modified approach to neural network detection of explicit or latent socio-cultural and cyber threats contained in the information content of Internet resources. Thematic dictionaries used for neural network classification of texts are pre-formed based on computer analysis of the target Internet content. An ensemble of combined ANN is constructed based on a combination of convolutional, fully connected and recursive layers, aimed at identifying text content containing undesirable information, including sociotechnical and cyber threats. The degree of recognition of thematic texts is reached in the range of 0.79…0.85. As a result, we obtained a set of neural network assessments of typical RSS feeds based on the criteria of detected signs of explicit or latent socio-cultural and cyber threats.