Bank transaction text label mining algorithms

Abstract

The banking transaction monitoring system implements decision support mechanisms for online payment control procedures for legal entities considering the dynamic risk profile of the client. The system includes a set of algorithms for the intellectual analysis of transaction parameters, including a text label for the purpose of payment, and decision support for an employee of the financial monitoring unit. The development of algorithms for analyzing textual labels for the purpose of payments allows us to clarify the dynamic payment profile of the user and increase the validity of the recommendations of the monitoring system. A block diagram of a system for identifying high-risk banking transactions based on data mining algorithms has been developed. Algorithms for data mining of textual labels of the payment purpose have been developed and the effectiveness of the proposed solution on field data has been evaluated. An algorithm is proposed for the phased analysis of the text label of the payment destination, including the stages of preprocessing, filtering, normalizing and constructing a classifier based on a set of regular expressions and intelligent analysis technologies. The difference between the algorithm is the use of adaptive category dictionaries and the multi-pass application of heterogeneous classifiers, which makes it possible to increase the validity of the decision on whether the transaction belongs to one of the selected classes.