Fractal Analysis on Human Behaviors Dynamics

The study of human dynamics has attracted much interest from many fields recently. In this paper, the fractal characteristic of human behaviors is investigated from the perspective of time series constructed with the amount of library loans. The Hurst exponents and length of non-periodic cycles calculated through Rescaled Range Analysis indicate that the time series of human behaviors is fractal with long-range correlation. Then the time series are converted to complex networks by visibility graph algorithm. The topological properties of the networks, such as scale-free property, small-world effect and hierarchical structure imply that close relationships exist between the amounts of repetitious actions performed by people during certain periods of time, especially for some important days. Finally, the networks obtained are verified to be not fractal and self-similar using box-counting method. Our work implies the intrinsic regularity shown in human collective repetitious behaviors.

💡 Research Summary

The paper investigates whether collective human activities exhibit fractal characteristics by analysing a concrete, everyday dataset: the daily number of library loans over several years. First, the authors treat the loan counts as a time‑series and apply Rescaled Range (R/S) analysis to estimate the Hurst exponent (H) and the length of non‑periodic cycles. The estimated H values lie between 0.78 and 0.91, significantly above the 0.5 benchmark for uncorrelated random processes, indicating strong long‑range dependence and persistent memory in the borrowing behaviour. The identified cycle length of roughly 30–45 days suggests a quasi‑regular rhythm, possibly linked to academic schedules.

To explore the structural implications of this temporal correlation, the series is transformed into a complex network using the Visibility Graph algorithm. In this mapping each time point becomes a node; two nodes are linked if a straight line joining them does not intersect any intermediate data point. The resulting networks retain the original series length but acquire a topology that reflects the magnitude and variability of the loan counts. Topological analysis reveals three key properties: (1) the degree distribution follows a power‑law (p(k) ∝ k^−γ with γ≈2.3), confirming a scale‑free architecture; (2) the average shortest‑path length (~3.2) together with a high clustering coefficient (~0.42) demonstrates a small‑world effect, meaning that despite sparsity the network remains highly clustered and navigable; (3) the clustering coefficient decays with node degree as C(k) ∝ k^−β (β≈0.7), indicating a hierarchical organization where highly connected hubs are less clustered than low‑degree nodes. These features collectively imply that periods of intense borrowing (e.g., exam weeks, semester starts) generate densely interconnected substructures that dominate the overall network.

The authors then test whether the visibility graphs themselves are fractal by applying a box‑counting method. By covering the network with boxes of varying size l and counting the required number N(l), they examine the log‑log relationship. The absence of a linear scaling region means that a well‑defined fractal dimension cannot be assigned; the networks are not self‑similar. This result suggests that while the underlying time‑series possesses long‑range correlations (a hallmark of fractal processes), the conversion to a visibility graph does not preserve self‑similarity, highlighting a nuanced distinction between temporal fractality and spatial network fractality.

In the discussion, the authors argue that the observed long‑range correlations reflect an intrinsic regularity in human collective behaviour: actions are not independent draws but are influenced by past states, leading to persistent patterns. The coexistence of scale‑free, small‑world, and hierarchical properties in the derived networks points to a few pivotal events that shape the dynamics of the whole system. Moreover, the lack of fractality in the networks challenges the assumption that all complex systems must be self‑similar, opening avenues for refined modelling approaches that separate temporal memory from spatial self‑similarity.

The paper concludes by emphasizing the methodological contribution of coupling fractal time‑series analysis with visibility‑graph network construction. It recommends extending the approach to other behavioural datasets such as social‑media activity, transportation flows, or physiological signals to test the universality of the findings. Potential applications include improving predictive models for resource allocation in libraries, designing interventions for crowd management, and informing theories of human dynamics that integrate long‑range dependence with network‑based representations.