Identifying Significant Task-Based Predictors of Emotion in Learning

摘要

Emphatic computing is concerned with enabling a system to recognize a user's current state and then providing the appropriate response to the user with the intention to support the user emotionally. However, in order to do so, the system must first identify the state of the user. Studies in computer-based tutoring are increasingly investigating ways to incorporate synthetic tutors that are equipped with computational models of empathy - in which these agents are trained to understand learners' emotions and respond based on the detected learner state. However, cultural differences affect the way people express and detect emotions. This paper attempts to identify the task-based features that could discriminate the learner's emotions in a Malaysian context. By studying several existing task-based features from literature, and combining them with new features, this study attempts to detect four frequent emotions that accompanies learning, namely, frustration, boredom, uncertainty and neutral. A user study is conducted with 33 students and results revealed that certain features can be used as predictors for the abovementioned emotions. Interestingly, results also showed that there is a tendency for students to choose synthetic tutors of the same race.