Move ordering and communities in complex networks describing the game of go

TL;DR

This paper models the game of go as complex directed networks of local patterns to analyze move sequences, community structures, and strategic properties, aiming to enhance computer simulations and understanding of human decision-making.

Contribution

It introduces a novel network-based approach to analyze go by constructing directed networks of local patterns and exploring their community structures.

Findings

Networks reveal groups of moves with shared strategic features

Community analysis distinguishes different game phases and player levels

Network approach can improve computer go simulations

Abstract

We analyze the game of go from the point of view of complex networks. We construct three different directed networks of increasing complexity, defining nodes as local patterns on plaquettes of increasing sizes, and links as actual successions of these patterns in databases of real games. We discuss the peculiarities of these networks compared to other types of networks. We explore the ranking vectors and community structure of the networks and show that this approach enables to extract groups of moves with common strategic properties. We also investigate different networks built from games with players of different levels or from different phases of the game. We discuss how the study of the community structure of these networks may help to improve the computer simulations of the game. More generally, we believe such studies may help to improve the understanding of human decision process.