NLP for education

Abstract

Natural Language Processing (NLP) is a form of Artificial Intelligence. It programs human language to a machine-readable format to achieve human-computer direct interaction. The importance of NLP-based virtual course assistants has raised its reputation to perform different education-related purposes. This paper proposes to develop an NLP-based virtual course assistant to provide educational support for both teachers and students. We believe adopting NLP-based virtual course assistant in class bring improvements in both teaching and learning experiences. With the help of an NLP-based virtual course assistant, it answers general and straightforward questions about the course 24/7. For example, students will ask the teacher many common questions, such as the weighting of each coursework, the time and place of the first lesson, etc. Yet, with a limited number of teaching assistants and teachers, answering similar questions to many students is time-consuming. It would be beneficial if the virtual course assistants could answer general questions about the course. So, the teaching team could devote more effort and time to answering complex questions. For students, an immediate response increases their motivation to study. It also avoids the embarrassment of directly approaching the teacher. The paper reviews four kinds of NLP tasks which are 1) Dialogue, 2) Question Answering (QA), 3) Question Generation, and 4) Sentiment Analysis (SA). Furthermore, the paper examines the performance of approaches to accomplish NLP tasks, including 1) Rule-based, 2) Chatbot Builder, and 3) Deep Learning.

The project scope defines as constructing a virtual course assistant for university students and teachers. The design of the NLP-based virtual course assistant needs to consider two main components: 1) Data and 2) NLP tasks. Firstly, there are no suitable datasets for the task proposed in this project. We need to gather different kinds of course data. Also, a sustainable method for further data collection is to develop a web-based course registration system for users to add and update course information. Appropriate access control and data integrity will need to concern while developing the web application. Secondly, to meet the need of teachers and students, a virtual course assistant will need to perform two kinds of NLP tasks, including QA, and SA. We adopt Bidirectional Encoder Representations from Transformers (BERT) as the key structure of models. We will fine-tune the models to meet different purposes.

The paper mentioned the performance evaluation design of the web-based course registration system and each NLP task performed by a virtual course assistant. To begin with, we will evaluate the web application according to a set of Website Performance Measurement metrics defined by Ghattas et al. (2020). Then, as some of our NLP tasks are closed domain, we will create our testing data to test the model and get the classification accuracy score. For the open domain SA task, we will test the model on the 100k-courseras-course-reviews-dataset to evaluate the model performance.