By Xiaoming Zhou and Cristina Conati
It has long been the focus of interest in the AI community to build intelligent machines, which behave logically and rationally. However, in recent years, people have begun to realize the importance of emotions in attention, planning, learning, memory, and decision-making, which got evidence from neuroscience, cognitive science, and psychology. Thus, several researchers in the AI community have begun to explore the issue of building "emotionally" intelligent computer, "a computer which is able to recognize and express emotions, respond intelligently to human emotions, and regulate and utilize its emotions".
One of the challenging problems of building emotionally intelligent computers is to recognize users' emotional states. We human use different sources of information to help us assess a person's emotional states, such as knowledge about that person's background, personality traits, contextual information, and also bodily expressions. But quite often we find that we are not good at this task and what that person is really feeling can be quite different from our assessment because the available information is often incomplete and sometimes even contradictory. Giving a computer the ability to recognize users' emotions is inevitably facing the same problem, which entails the task a high level of uncertainty.
We propose a general framework for affective student modeling in the environment of an electronic educational game, Prime Climb. This framework is based on an existing cognitive theory of emotions and uses Dynamic Decision Networks to combine different sources of information to deal with the high level of uncertainty involved in this modeling task. We apply the framework to build an affective user model to be used by intelligent pedagogical agents in an electronic educational game.
After describing an initial model that is based on two preliminary studies, we present the results from a formal Prime Climb user study and the refined model based on the result.