Fine-Tuned BERT Enabled Context Aware Virtual Learning Assessment Model

Abstract

This paper presents a state of art approach to perform pre-assessment of learner through the feedback in a Virtual Learning Environment (VLE). Student Reviews presents the context, their sentiments and suggestions about the course. These contexts are modelled using Bi- Directional Encoder Representations from Transformers (BERT). BERT performs the Natural Language Processing (NLP) tasks. It has ability to understand the reviews and do classification. It can predict early that “if student can “Pass” or will “fail” in the examination based on their reviews about the course. This can help in generating alarm for instructors to focus students differently. This paper presents the experiment procedures and results. More than 200K unique reviews are assessed with Fine Tuned BERT and with 95% accuracy has been achieved. This work can add additional dimension to the conventional assessment method. A self-evaluating deep learning model can help the instructors and make them understand about the learner’s sentiment after each assessment. Hence brings more quality to teaching learning process.