Uma metodologia para modelagem de jogadores baseada em aprendizado de máquina

Abstract

Artificial Intelligence (AI) is gradually receiving more attention as a fundamental feature to increase the immersion in digital games. Among the several AI approaches, player modeling is becoming an important one. The main idea is to understand and model the player characteristics and behaviors in order to develop a better AI. It is possible to model player aspects in different levels of abstraction, such as actions, position, preferences, knowledge and satisfaction. This modeling allows games to customize their AI, difficulty or levels to specific players, making the game experience more interesting. In this work, we discuss several aspects of this new field. Since several works have been tackling this problem, we proposed a taxonomy to organize the area, discussing several facets of this topic, ranging from implementation decisions up to what a model attempts to describe. We then classify, in our taxonomy, some of the most important works in this field. Besides the taxonomy, we also presented a generic approach to deal with player modeling using machine learning, and we instantiated this approach to model players' preferences in the game Civilization IV. The instantiation of this approach has several steps. We first discuss a generic representation, regardless of what is being modeled, and evaluate it performing experiments with FPS and strategy games (Counter Strike and Civilization IV, respectively). Results show the effectiveness of this representation in characterizing and modeling agents. By observing matches we were able to infer agents' models; and varying agent's models we were able generate different behaviors.Continuing the instantiation of the proposed approach we evaluated the applicability of using game score information to distinguish different preferences. To perform this task we presented a characterization of virtual agents in the game, comparing their behavior with their stated preferences. Once we have characterized these agents, we were able to observe that different preferences generate different behaviors, measured by several game indicators. Using this information we tackled the preference modeling problem as a binary classification task, with a supervised learning approach. We compared four different methods, based on different paradigms (SVM, AdaBoost, NaiveBayes and JRip), evaluating them on a set of matches played by different virtual agents. We obtained accuracies that improved by far the state of the art. We conclude our work using the learned models to infer human players' preferences. Using some of the evaluated classifiers we obtained accuracies over 60% for most of the inferred preferences.