University of Alicante (SPAIN)
About this paper:
Appears in: EDULEARN22 Proceedings
Publication year: 2022
Pages: 8707-8714
ISBN: 978-84-09-42484-9
ISSN: 2340-1117
doi: 10.21125/edulearn.2022.2077
Conference name: 14th International Conference on Education and New Learning Technologies
Dates: 4-6 July, 2022
Location: Palma, Spain
This paper describes a methodology of a framework for the design and practical implementation of chatbots that respond to questions from students interacting through virtual tutorials. The methodology is validated through consultation to experts, specifically to Theory and Practices professors in the subject of Information Technology and Systems, at the Computer Science degree of the University of Alicante (UA). The proposal stems from the main need to respond, in a short period of time, to the numerous tutorials generated by groups with a large number of students enrolled. A question arises on how to assist both the teachers and the students with automatic tutoring tools that are capable of interpreting natural language and generating high-quality responses, with the ability to pass the Turing Test, while complying with the educational objectives and instructions in the curricular program. The implementation of virtual tutors based on RASA technology is described, based on practical experiences during the teaching of the e-Learning subject in the Multimedia Engineering degree, at the UA. The technological solution is based on a conversational training data corpus created by students who develop different projects in the Project-Based Learning (PBL) modality. The solution allows the development of a generative chatbot, capable of combining a broad conversational domain with specific topics based on rules. The proposal is based on neural network models for Natural Language Interpretation, through either voice or text, and Dialogue Management through inference. The workflow can be customized with different tools for tasks such as Word Embedding, Name Entities Recognition, Regular Expressions and Tokenizers. We describe how to reuse the dialogues generated by the students through Transfer Learning. The solution can identify the conversational context based on the conversation’s natural flow with the assistant, as well as detecting key aspects for preparing responses based on the identification of the user's intention. The models are based on Transformer-type networks, capable of incorporating advanced self-attention mechanisms. By managing “chit-chat” type conversations, it is possible to provide the conversation with a natural aspect, with the goal of keeping the student connected throughout the tutorial and avoiding early dropouts. The multi-channel perspective and ubiquitous features are offered through the integration of the solution into various communication channels such as Slack, Instagram or a web page. The case study is extended to the Information Technology Systems course in the Computer Science degree. Conclusions show the main results of integrating Data Analysis and Visualization techniques used during the creation of the Training Corpus, with others from Content Generation and Natural Language Processing.
Chatbots, PLN, tutoring system, artificial intellegence.