An Investigation On Human Dynamics Based On Email Networks

Human dynamics offers a breach for the quantitative analysis of human behavior's statistical rules. It gets widely-spread attention because of the importance of its theory and application. However, currently the research of human dynamics is not mature yet, and different human dynamic models give un-unified explanations for the non-Poisson attributions of human behaviors. To analyze and examine the scope of the application and rationality of the conclusions for existed models, this article analyzes the email communication patterns of users reflected in a national university's email network. It is shown that the two existed models, priority-based queuing model and the cascading non-homogeneous Poisson process model, both fail to describe properly the dynamic mechanism in the email network. The hypothesis in the priority-based queuing model that users share a universal behavior pattern is unreasonable; besides the cascading non-homogeneous Poisson process cannot embody the unexpected cessation or dynamic evolution in individual activities. Therefore, the clustering-analysis-based segmented cascading non-homogeneous Poisson process model is promoted which can produce better predictions.The main contributions are summarized as follows.1. Review the development history of human dynamics for more than ten years, including the discovery of new statistical characters of human behaviors, the present situation, bottle-necks and existed problems in national and overseas research. Also discuss the principle and conclusions for respective model specifically.2. In use of the log data of the Email network in a national college, analyze the priority-queue model and cascading non-homogeneous Poisson process model in detail. Examine the assumptions and simulation results strictly through hypothesis testing for both models. Based on the results from the examination, probe deeply the scope of application of the models and the reasons they can't reappear the non-Poisson character of human behaviors in the real network.3. On the basis of the previous conclusions, propose a method to apply the cluster-analysis methods in the human dynamics modeling, producing the clustering-analysis-based cascading non-homogeneous Poisson process model. This model divides a long time series into several relatively more stable short ones through cluster analysis. Therefore it could eliminate the effect of individual non-regular behaviors on modeling, improve the simulation accuracy and make the simulated results more close to the actual human behaviors'characters displayed in the real network.