Problems of Analyzing Socio-Political Content of Internet Resources Based on Neural Network Technologies

Ensuring information security requires identifying undesirable information content of Internet resources. Semantic and lexicological diversity of Internet content requires improvement of methods of neural network analysis of natural language texts (NLP). The problem is complicated by the presence of” information garbage”, which is a specific information noise that complicates the task of classifying texts. Well-known NLP technologies using artificial neural networks (ANN) include substantiation of the structure and construction of a subject-oriented database of text data bodies, frequency analysis and construction of dictionaries. To identify semantic content and latent threats, a dense vector representation of the analyzed texts in a multidimensional space (embedding) is justified. The authors substantiate a modified NLP approach to identifying sociocultural and cyber threats, contained in the information content of Internet resources. To justify and research the ANN architecture and hyperparameters focused on the socio-political content under study, the ANN family was built in Python 3. The ANN architecture included combinations of fully connected, convolutional, and/or recurrent layers. The number of neurons of the recognizing fully connected layer with the “softmax [[CHECK_DOUBLEQUOT_ENT]] activation function (or sigmoid in multiple classification) was taken by the number of classes marked in the text corpus.