ALGORITHMS FOR CATEGORIZATION AND TEMPORAL ANALYSIS OF BOT-PROGRAMS IN SOCIAL NETWORKS

Abstract

Recent studies confirm the growing share of bot-driven activity in digital communication environments. According to Imperva's 2024 report, nearly half of web traffic is generated by automated agents, with over 30% attributed to malicious bot-programs. The accessibility of tools based on generative AI has contributed to this growth, enabling even non-specialists to launch automated scraping and spam campaigns. Simultaneously, Cloudflare data show that bot-programs target critical areas such as finance and e-commerce, causing measurable harm to businesses and undermining user trust. Academic research highlights both the increasing sophistication of bot-programs and the emergence of new detection strategies, including behavior-based modeling and machine learning methods. However, many existing approaches remain limited in their ability to integrate structural and temporal dimensions of user activity.

The study aims to analyze behavioral characteristics of users in digital communication platforms, with a focus on the distinction between human-operated and automated accounts. The goal is to trace the distribution and evolution of various account types across time and thematic sectors, as well as to identify sectors most vulnerable to the influence of coordinated bot-program activity.

To achieve this, the research introduces two purpose-built algorithms: one for user categorization and another for temporal activity analysis. The categorization algorithm accommodates both pre-labeled and raw user data, using heuristic thresholds and public bot-likelihood metrics to assign accounts to one of three groups: humans, benign bot-programs, and malicious bot-programs. The temporal analysis algorithm models user activity over time by comparing expected versus actual interaction levels per group, enabling the identification of overstimulated or understimulated behavior patterns. These algorithms were applied to a dataset comprising the TwiBot-2022 benchmark and a manually compiled corpus of tweets from 2023. The resulting analysis tracks the changing distribution of user types between 2017 and 2023 and maps their activity across multiple thematic sectors.

The findings reveal a noticeable rise in malicious bot-programs activity between late 2019 and 2020, coinciding with the outbreak of the COVID-19 pandemic and the U.S. presidential election—both of which are known triggers for disinformation campaigns. Furthermore, sectoral analysis shows disproportionate bot-programs involvement in domains such as politics, marketing, and financial services. Meanwhile, categories such as entertainment and lifestyle remain dominated by human activity. The study demonstrates the effectiveness of the proposed algorithms in capturing dynamic and structural aspects of online behavior and contributes new tools for the monitoring of digital ecosystems and the mitigation of information threats.