Modeling bursts and heavy tails in human dynamics.

Area of Science:

• Complex Systems Science
• Statistical Physics
• Human Dynamics

Background:

• Understanding human behavior is crucial for modeling social, technological, and economic systems.
• Existing models often assume human actions are randomly distributed in time (Poisson processes).
• This assumption fails to capture the observed bursty nature of human activities.

Purpose of the Study:

• To provide empirical evidence that human actions exhibit non-Poisson statistics.
• To identify the underlying mechanisms driving bursty human behavior.
• To propose and validate queuing models that accurately describe human activity patterns.

Main Methods:

• Analysis of five distinct human activity patterns: email, letter communication, web browsing, library visits, and stock trading.
• Statistical analysis of inter-event times to identify deviations from Poisson distributions.
• Development and testing of two queuing models: one with unlimited task capacity and one with limited queue length.

Main Results:

• Human activity patterns demonstrate non-Poisson statistics, characterized by bursts of activity followed by inactivity.
• This burstiness is explained by decision-based queuing processes where tasks are prioritized.
• Empirical data supports two queuing models: alpha=3/2 for unlimited capacity (e.g., surface mail) and alpha=1 for limited capacity (e.g., email, web browsing, library visits).

Conclusions:

• Human behavior is inherently bursty, deviating significantly from random (Poisson) distributions.
• Queuing processes, influenced by task prioritization, are the primary drivers of this burstiness.
• The findings necessitate revised models for human dynamics, with implications for risk assessment, communication, and economic forecasting.