Learning to Trade in Strategic Board Games

Automated agents in multiplayer board games often need to trade resources with their opponents-and trading strategically can lead to higher winning rates. While rule-based agents can be used for such a purpose, here we opt for a data-driven approach based on examples from human players for automatic trading in the game "Settlers of Catan". Our experiments are based on data collected from human players trading in text-based Natural Language. We compare the performance of Bayesian Networks, Conditional Random Fields, and Random Forests in the task of ranking trading offers, and evaluate them both in an offline setting and online while playing the game against a rule-based baseline. Experimental results show that agents trained from data from average human players can outperform rule-based trading behavior, and that the Random Forest model achieves the best results.