Description
Date depot: 18 novembre 2021
Titre: Sentiment Analysis on Online Learning: Applying Ensemble Methods on Social Networks Students Reviews to assess online education
Directeur de thèse:
Lionel TROJMAN (LISITE)
Encadrant :
Rafael ANGARITA AROCHA (LIP6)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Traitement automatique des langues et de la parole
Resumé: The drastic change in digitalizing learning showed challenges, strengths, and weaknesses from instructors' and students' perspectives. From the mental and emotional side, online learning helps students to become more independent learners, yet research shows that students still prefer classroom learning for diverse reasons.
Nowadays, students of different ages and different domains express their thoughts, comments, and interactions on social networks. Sentiment Analysis using reviews of student evaluations over social media can provide information that determines the performance and assessment of instructors and helps to improve the level of student satisfaction in recent learning environments. For this purpose, several tools and techniques are used to extract the insight of social media data, break it into parts, give weights for each, identify emotions and opinions, and classify texts into positive, negative, or neutral using sentiment libraries.
The accuracy of classification methods detecting the sentiment in the texts varied between supervised and unsupervised classification algorithms. However, ensemble methods in all comparisons showed higher accuracy. The work in [6] combines Naive Bayes classifier, Random Forest classifier, SVMs, and Logistic Regression, versus single classification algorithms for sentiment classification of a dataset of Twitter texts. The proposed method showed higher performance and accuracy. Another study [7] for classifying another set of Twitter texts showed that ensemble techniques, including the use of Natural Learning Process Techniques (NLP), performed an increase of accuracy with 3-5%.
To cope with these difficulties and specifically the low accuracy, it is necessary to find the best performing ensemble of classifiers for Sentiment Analysis for student texts over different social media applications and their evaluation for online education.
Doctorant.e: Al Katat Souha