CASPIAN JOURNAL

MANAGEMENT AND HIGH TECHNOLOGIES

METHODOLOGY OF SOCIAL ENGINEERING ATTACK DETECTION BASED ON NATURAL LANGUAGE ANALYSIS ALGORITHMS

Read

Chastikova Vera A., Gulyai Victoria G. METHODOLOGY OF SOCIAL ENGINEERING ATTACK DETECTION BASED ON NATURAL LANGUAGE ANALYSIS ALGORITHMS // Caspian journal : management and high technologies. — 2022. — №3. — pp. 61-71.

Chastikova Vera A. - Kuban State Technological University

Gulyai Victoria G. - Kuban State Technological University

Recently, along with technical attacks on users of various electronic communication systems, the number of social engineering attacks has increased. So, over the past year, an increase in social engineering attacks has been recorded by more than 90 %. Currently existing ready-made products from various manufacturers are not able to fully combat attacks of this type. In this paper, we consider a new approach to solving problems of detecting social engineering attacks - the use of natural language analysis algorithms. In order to experimentally test the possibility of using these methods within the framework of the task, the following algorithms were implemented: the bag-of-words language model, the Word2Vec word embedding algorithm and the BERT method based on the Transformer architecture. According to the results of the study, it was found that the best results were shown by the BERT model, in which the accuracy of processing data from the control sample was 97.35 %. It is also worth noting the bag-of-words algorithm, which has a significant advantage over other models in data processing speed - approximately 1-2 ms per data processing epoch. The Word2Vec algorithm showed average results relative to the bag-of-words and BERT models. This word embedding algorithm has a processing accuracy advantage over the bag-of-words language model, and a processing speed advantage over the BERT algorithm.

Key words: social engineering, natural language processing algorithms, bag-of-words, Word2Vec, BERT