A Novel Approach to Automated Content Generation for Education using GPT-3

Anonymous

Abstract:

The primary objective of this research is to investigate automatic content generation in an educational context. In an era characterized by an unprecedented influx of information, the conventional methods of content creation for classroom instruction have been rendered increasingly inadequate, thus the motivation behind this research is to aid teachers in generating content for educational use such that they won’t need to expend much time and energy as with traditional methods.

Modern methods of generating content for the classroom are sought after due to the benefits when compared with more traditional methods. One example of this is a case study carried out amongst 48 college students where a positive effect occurred in the students’ learning outcomes when they used computer-generated questions.

With automated content generation being the primary focus of this research, this research heavily relies on and investigates Natural Language Processing (NLP) techniques and technologies. Thus we delve into how automated content generation for previous systems was carried out along with Large Language Models (LLMs)

Our methodology relies on making use of the GPT model, GPT-3, the proposed system performs various NLP tasks such as Summarization and Information Retrieval (IR) along with prompt engineering to generate content within an educational context and empower educators when it comes to generating content. The system accepts inputs from the user that may be plain text, a YouTube video or a PDF and then generates content, such as a worksheet with questions in return by interfacing with and using GPT-3 to generate the content.

One also must keep in mind that such a system raises ethical qualms, particularly regarding data privacy and bias. Algorithmic bias is a commonly known issue within the field of NLP, as bias often arises from biased training data and algorithms. This bias can be harmful as it can directly affect the learning outcomes of certain groups of students. Furthermore, as such a system may collect learner data, data privacy comes into question, particularly who or what has access to this data and how it is used. A limitation of the currently proposed system is that as it uses GPT-3 as a backend, it will incorporate the same bias as GPT-3. The system however does not pose a data privacy risk as no sensitive or personal information is asked for, and the given inputs are only retained up until the corresponding output is generated.

In conclusion, this research focuses on the integration of computational linguistics within the field of education through the integration of GPT-3 with the application of automated content generation. The results of this study show a positive trend as 94\% of the respondents said that the system generated relevant content while 85\% of respondents said that they would adopt such a system.

This work raises the question of how NLP can be utilised more effectively within the field of education. Furthermore, this system, while currently aimed at primary and secondary level students at a general level, in future work it can potentially be adapted for particular grade levels and particular topics by fine-tuning the model.

Paper Type: long

Research Area: Generation

Contribution Types: Approaches to low-resource settings, Surveys

Languages Studied: English

Consent To Share Submission Details: On behalf of all authors, we agree to the terms above to share our submission details.