Data Mining of B2b Transactions Using a High-utility Temporal Approach

Modern B2B e-commerce systems generate vast arrays of transactional data, yet classical association rule mining methods face several critical limitations, notably the "financial blindness" of binary models and the "temporal blindness" of static approaches. These constraints hinder the discovery of patterns with genuine economic value and overlook the cyclical nature of wholesale procurement.Methods. This paper proposes a high-utility temporal approach to B2B transaction analysis. The approach is centered on the UP-Growth (Utility Pattern Growth) mathematical model, adapted to operate with a utility function based on the rational transaction context, defined as the product of individual price and wholesale purchase volume. To address the anti-monotonicity problem, the concept of transaction-weighted utility is introduced. The integration of the temporal factor is achieved through the analysis of inter-purchase interval vectors, utilizing the coefficient of variation as a criterion for cycle stability. As a result, a holistic dual-stage pipeline architecture for the intelligent analysis of B2B customer transactional data has been developed. In the first stage, the pipeline performs the extraction of High-Utility Itemsets, while the second stage involves calculating the trigger date to establish a proactive offer window. A comparative analysis with the classical FP-Growth algorithm confirmed the superior efficiency of the proposed approach in identifying economically significant patterns, validated using real-world data from the "Baza Vzuttya".