Predicting variable-length paths in networked systems using multi-order generative models.

Area of Science:

• Complex Systems Science
• Network Science
• Data Science

Background:

• Networked systems generate path data, which are temporally ordered sequences of nodes constrained by topology.
• Understanding path patterns is crucial for analyzing complex systems and optimizing engineered systems like supply chains and mobility services.

Purpose of the Study:

• To introduce MOGen, a generative modeling framework for accurate next-element and out-of-sample prediction of paths.
• To develop a parameter-free model selection approach that automatically identifies the optimal model from data.
• To establish a mathematical formalism linking path models to random walks in multi-layer networks.

Main Methods:

• Developed MOGen, a generative modeling framework for path prediction.
• Implemented an automated model selection approach for parameter-free optimization.
• Introduced a mathematical framework connecting higher-order path models to multi-layer network random walks.

Main Results:

• MOGen demonstrates high accuracy and consistency in both next-element and out-of-sample path prediction.
• Empirical data shows MOGen surpasses current state-of-the-art sequence modeling techniques.
• A novel mathematical formalism was established linking path models to random walks in multi-layer networks.

Conclusions:

• MOGen provides an effective and automated solution for modeling and predicting paths in complex networked systems.
• The framework advances the understanding of structure and dynamics in complex systems through accurate path analysis.
• The developed mathematical formalism offers new insights into the relationship between path data and network dynamics.