Automatic Student Affective State Detection from Plain Text

Abstract

We explore the concept of automatic detection of affective state of a learner in an e learning environment. We propose a model to detect the emotion from learners’ text. We employ machine learning algorithms with ISEAR data and twitter data from Kaggle data repository. We follow the conventional steps of natural language processing; text preparation, feature extraction and emotion detection and classification. For text preparation we use processes of tokenization and segmentation, noise removal and segmentation. We extract features using count vectors and term frequency -inverse document frequency. For classification compare varied machine leaning algorithms. Results show that Linear SVM using Count Vectors accuracy gave an accuracy of 79% which is encouraging. We deduce that we can extract the affective states of the learners automatically from text during their interaction with e-learning environment. This will help in understanding the learners needs and help in enhancing adaptability