Аннотация
This paper proposes an entropy-normalized multidimensional framework for user activity segmentation in the Reddit social network. The relevance of the study is determined by the rapid growth of large-scale weakly structured behavioral data generated by modern digital platforms and online social systems. User engagement indicators such as likes and comments are shown to exhibit heavy-tailed statistical distributions characterized by strong variability, nonlinear interaction dynamics, and scale-free structural properties. Such characteristics significantly limit the applicability of classical statistical assumptions and motivate the development of scalable analytical approaches for intelligent behavioral modeling. To address this problem, we introduce a composite activity index integrating standardized engagement indicators with normalized Shannon entropy of temporal behavior. The proposed representation enables simultaneous analysis of interaction intensity and temporal regularity within a unified multidimensional behavioral framework. Temporal entropy is used to quantify uncertainty and irregularity of user activity across hourly intervals, allowing the model to distinguish structurally different behavioral profiles even when engagement intensity remains similar. Heavy-tailed properties of engagement distributions are investigated using power-law analysis and complementary cumulative distribution function (CCDF) modeling in logarithmic coordinates. User segmentation is performed using the KMeans clustering algorithm, while clustering quality is evaluated using the silhouette coefficient. Experimental analysis confirms the existence of scale-free interaction dynamics in Reddit engagement data and demonstrates that entropy-normalized representations improve cluster separation and interpretability compared to raw-feature clustering approaches. The developed analytical pipeline is computationally scalable and suitable for large-scale digital platforms containing massive volumes of user-generated interactions. The proposed framework may be applied in intelligent analytics, recommendation systems, anomaly detection, behavioral profiling, and computational social science research. Furthermore, the obtained results indicate that entropy-aware multidimensional representations provide an effective foundation for future graph-based interaction analysis and machine learning approaches in complex online ecosystems.