USING MACHINE LEARNING IN REAL-TIME BIDDING

Abstract

The relevance of the study is stipulated by the need for automated and prompt decision-making in real-time bidding systems (RTB) operating in the dynamic environment of digital advertising based on a large number of user and contextual parameters. Given the growth of data volumes and high requirements for content personalization, there is a need to implement more flexible and adaptive algorithms. The purpose of the article is to study the potential of integrating machine learning (ML) methods into the processes of managing advertising rates in order to increase the relevance of ads and the effectiveness of advertising campaigns in an unstable market environment.

The study applies structural-functional, analytical and comparative approaches to studying the architecture of RTB systems, as well as modeling algorithms for real-time recommendation decisions. Particular attention is paid to such factors as input data quality, model update frequency, system adaptability, and application programming interface (API) integration. It is found that the most commonly used methods in practice are logistic regression, gradient boosting, deep neural networks, and reinforcement learning (reinforcement learning is a method in which a model is trained through successive decision-making with reward or punishment). It is proved that the effectiveness of advertising strategies increases significantly when using such indicators as key performance indicators (KPIs) and customer lifetime value (LTV).

The main limitations of ML implementation in RTB systems are identified: insufficient computing resources, low interpretability of algorithms, and risks of over-automation. The expediency of applying hybrid models with a combination of offline and online analysis, as well as the use of explainable AI to increase the transparency of decision-making is substantiated. The scientific novelty lies in the systematization of factors that critically affect the accuracy of forecasts in a dynamic advertising environment. The prospect of further research is the creation of models that are resistant to the limitations of personalized data and capable of context-dependent self-updating.