Building a Vector Model of Textual Information to Analyze the Criteria for Identifying Illegal Sites Offering Microfinance Services

Today, there is a rapid increase in the number of fraudulent transactions in the field of financial services to the population, in particular, Internet payments as well as fraudulent Internet resources represent a large share where the task of identifying illegal websites offering microfinance services is most urgent. In this paper, the proposed criterion for identifying fraudulent sites by analyzing the unstructured features of web resources that characterize the content of suspicious web resources has been studied. The characteristics are based on textual information collected from the researched online resources. In this paper, a comparative analysis of the models of the vector representation of text data, namely, the model based on the matrix of Positive Pointwise Mutual Information, Word2Vec, Doc2Vec, is carried out. After that, a binary classifier of the received representations of text data using Support Vector Machine was built. To confirm the effectiveness of the constructed model, cross-validation was used, as well as classification quality metrics were calculated. The built algorithm detects fraudulent web resources offering microfinance services based on text information with an accuracy of 98.3%, which indicates the effectiveness of the proposed method.