University of Malta (MALTA)
About this paper:
Appears in: ICERI2021 Proceedings
Publication year: 2021
Pages: 4192-4201
ISBN: 978-84-09-34549-6
ISSN: 2340-1095
doi: 10.21125/iceri.2021.0978
Conference name: 14th annual International Conference of Education, Research and Innovation
Dates: 8-9 November, 2021
Location: Online Conference
In the classical classroom setting, an educator is tasked with transferring knowledge to multiple students. An immediately noticeable flaw, however, is that when learning, students may need additional attention, however, that is not always possible when there is a class of twenty-five students but only one educator. Our Educational Chatbot system makes use of AI (Artificial Intelligence) to engage in a dialogue with the students and help them learn Mathematics by providing explanations similar to those provided by the educator.

The novelty of our work lies in the fact that we are moving away from traditional systems. In fact, such systems tend to fall into two main categories. The first makes use of hard-coded responses which are handcrafted specifically for this task. The second uses dynamic responses which are still built upon a well-defined script.

Our chatbot uses a hybrid approach. The underlying responses or scripts are not hardcoded. Our AI uses the explanations which the educator provides to the children and generalizes over them using the T5 transformers model and the GPT2 dataset. So in essence, no special format is required to teach the AI since it is capable of processing the notes and create dynamic responses out of them automatically. In so doing, the educator or the AI expert does not need to set up the system since it is done automatically.

We tested the system with around 100 people that are involved in education and the results are very promising. In general, the system was well received with 73% of respondents indicating that they enjoyed using the chatbot. Furthermore, 77% of them claimed that the chatbot helped them understand Mathematics better. Although the system is far from being perfect, the results are very positive and encourage us to keep on improving our automated hybrid approach.
Natural Language Processing, Text Generation, Artificial Intelligence Assisted Learning.