The Study on Recognizing Learning Emotion with the Features of Facial Expression

• Main Authors: Li-Chun Sue, 蘇莉鈞
• Other Authors: Kuan-Cheng Lin
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/56568138331345709106

碩士 === 國立中興大學 === 資訊管理學系所 === 105 === Facial expression is the ways of communication for the human, expressing their emotions by the facial expression. In the learning activities, the learners'' emotional expression will directly or indirectly affect their learning motivation and effectiveness. The change of emotions for the learners in the learning process is complex and diverse, and the emotions are mainly to explore the emotions about learning, which are: Cognitive Emotions and Academic Emotions. Cognitive emotions are complex. The label of Cognitive emotions not only can represent a variety of emotions for the single label but also is numerous names which are often discussed are: Confusion, Engaged and so on. Cognitive emotions are usually the short-term emotions. Hence, the academic emotions are referred to the learners’ feeling which through the cognitive assessment in the learning environment. It is integration and review of emotions for the learners in the leaning process. However, the domestic and foreign researches in recent years to obtain academic emotions are mostly filled "Achievement emotions questionnaire (AEQ)". This study summarizes the Cognitive Emotions, which are emotions about learning and corresponds to Academic Emotions. In order to name a new set of emotions: Learning Emotion, which is a shorter duration of Academic Emotion. And the classes (or labels) of Learning Emotion are same as Academic Emotion: Enjoyment, Hope, Pride, Boredom, Anxiety and also have a Neutral expression. The purpose of this study is to establish an effective learning emotions prediction model because we want to know the learners’ academic emotion without filling the questionnaire and by observing the sequence of learning emotion we can realize the leaners’ emotional change which caused by leaning activities in the learning process. Therefore, this study will establish the database about Learning Emotion, named as “Learning Emotion Face Image Database”, as the training data for the model. Under the image processing, two different facial features in the image are captured, which are “Feature Value” and “Action Units”. These facial features are input vectors for machine learning approaches as SVM, MLP and RF. The best performance of above-mentioned set: Feature Value and RF will be chose as our model’s method. Then through the process of feature selection and model parameter optimization, the accuracy rate of prediction model is over 87% for training data in 80/20 rule. In this study, the results of prediction model: Learning Emotion States are compared with the outcome of AEQ: Academic Emotion States. The averaging correct rate of learning emotion states to academic emotion states is 60.47%. Nevertheless, after enlarging twice the latter part of learning emotion sequence, the highest correct rate is 77%. This shows that the latter part of the learning emotion sequence for detecting academic emotions is more influential than the previous part of learning emotion sequence. As the result of high correct rate, we verify our prediction model in this study has a certain degree of trustworthiness.