Abstract
Creating believable social behaviour is a challenging task. We propose that imitating human players’ actions and having into account the influence of the players’ social identity improves the decision making process of an artificial agent, in terms of believability. To test this hypothesis, a predictive model was developed, based on a probabilistic approach to plan recognition using Dynamic Bayesian Networks (DBN) in both learning and testing phases. This work has been developed in the context of the INVITE game and aims to create an artificial agent described above and test if it is believable in games with human players.