Predicting Group Behavior from Profiles and Stereotypes

摘要

Behavior prediction is the underlying component of much recent modeling and analysis efforts, and the underlying assumption is that such prediction is possible, via some form of computation. The problem is inherently difficult, challenging theoretical ("What are the theoretical limits of behavior prediction? Which factors are the best indicators of future behavior?"), practical ("Does the necessary knowledge about individuals and groups exist?") and pragmatic ("Can such knowledge be collected?") considerations. In general, it is easier to predict the behavior of groups than of individuals. In this paper, we describe a knowledge-based approach to group behavior prediction that integrates a number of knowledge sources and inferencing approaches within a computerized decision-aid, designed to function in data-sparse environments. The core knowledge components of the decision aid are: 1) a group profile, consisting of the most critical behavior determinants, derived from theoretical and practical knowledge; 2) library of group stereotypes, and 3) knowledge relating profile and stereotype components to support the derivation of additional information from the available data, including likely group behavior. In the current implementation, the decision-aid is designed to support a PSYOP analyst and focuses on predicting the likelihood of group violence (including the most likely means and targeting). However, we believe that the knowledge templates and inferencing methods, as well as associated visualizations, are applicable for behavior prediction in other domains.