Modeling crowd behavior

Modeling crowd behavior is an important challenge for cognitive modelers, multi-agent systems and social simulation. In this dissertation we explore several techniques for modeling crowd behaviors and address important challenges concerning each technique. In the first part of the work we use the agent based approach for modeling crowd behavior. We present the extended Social Comparison model (SCT), which we believe is a general cognitive process underlying the social behavior of each individual in a crowd. In this work we provide a qualitative evaluation of the SCT model, as well as others, in contrast to human pedestrian behavior. The results clearly demonstrate that the SCT model is superior to others in its fidelity to human pedestrian behavior. Moreover, we have focused on an open question which has arisen from the SCT model, namely when should the SCT process be used at the architectural level in order to guide action-selection of agents. We have extended the SCT model to address this open question. We argue that comparisons take place all the time (i.e., differences are perceived and processed), but the cognitive architecture limits actions taken to minimize differences in cases where the comparisons yield significant differences. In addition we examine the impact of cultural differences on the macro level behavior produced in pedestrian and evacuation domains. In this work we have advanced by treating culture as a first-class object in models of physical crowds and have extended the SCT model accordingly. We introduced cultural individual-level parameters into the simulations, and then examined the effects of these individual level parameters on the emergent crowd dynamics. In the second part of the work we use the qualitative reasoning approach for modeling demonstrations. We present a first attempt to use qualitative reasoning techniques in order to model crowd behaviors. To the best of our knowledge, such techniques have never been applied to modeling and reasoning regarding crowd behaviors, nor in particular demonstrations. We have developed qualitative models consistent with the partial, qualitative social science literature, which has enabled us to model the interactions between different factors that influence violence in demonstrations. We then utilized the qualitative simulations to predict the potential eruption of violence, at various levels, based on a description of the demographics, environmental settings, and police responses. In addition to providing predictions, the resulting qualitative simulation graphs were analyzed to determine the factors