A machine learning approach to classify student Emotions from text (T-0743) (MFN 6911)

Abstract

An Intelligent Tutoring System (ITS) is a software that is established on Artificial Intelligence (AI) and Machine Learning (ML) automations, devised to transfer educational material and give feedback to students. The core target of an ITS, is to adapt to a student's learning requirement like an expert human teacher. Students are provided an environment wherein they can ask questions and give answers about tasks of varying complexities, and exercises with persuasive feedback given by an ITS. Emotions playa very important role in any form of interaction and it is an imperative for an ITS to be able to react to a students' emotional states during an interaction. The research aims to make an ITS meta-emotionally intelligent by enhancing the Learner Model. For this, we have proposed a proof-of-concept meta-emotional state model i.e. sub component of the Learner Model that classifies students' emotions by using Common Student Utterances (CSU) collected from an ITS: NDLtutor. We added general utterances from social media and educational blogs in order to expand the volume of the data set. We implemented two supervised ML approaches, i.e., Multinomial Narve Bayes (MNB) and Stochastic Gradient Descent (SGD) algorithm to classify emotions against CSU. The MNB and SGD . models are simple and smart ML algorithms that show realistic hypothetical results. We conducted the experiment upon 70 students of Bahria University to analyze the model proficiency. The experiment provided encouraging results with our classifiers predicting the emotions with an accuracy of -77%. The precision of the model can competently be improved by expanding the size of the data set. Such a classification scheme can allow an ITS to be more adaptive to the changing emotional states of a student during an interaction.