A Multi-Lingual Conversational AI Chatbot for Effective Educational Consultations: A Case Study of ACE-DS, University of Rwanda

Abstract

The requests for real-time simultaneous direct client-staff interactions and consultations in most industries and organizations is on the increase, particularly in the educational sector. These concurrent and numerous requests have led to queues, turmoil, and service delays, especially during and after working hours due to the shortage of employees and language barriers to cater for the non-stop services needed by clients. Artificial intelligence (AI), Natural Language Processing (NLP) and linguistics technologies have become a potent tool for creating innovative solutions to tackle these challenges. To this end, this study aims to better understand question-answering systems, with an emphasis on creating a multilingual conversational AI chatbot that will ultimately offer a workable solution in an educational setting. The multi-lingual chatbot system developed in this research is designed to answer various consultation queries related to ACE-DS services. This covers tuition fees, admission procedures, postgraduate programs, facilities, program application requirements, program modules, travelling requirements for international students, student accommodations and so on. The study used annotated Frequently Asked Questions (FAQs) and corresponding answers about ACE-DS, incorporating sentiment and conversation data to train a Deep Learning (GRU RNN) algorithm. This enables the chatbot to engage in meaningful conversations, detect user moods, and provide appropriate responses based on its prediction. Additionally, the chatbot system incorporates language detection and translation capabilities, allowing users to engage the system in multi-lingual conversation. The results of the system evaluation and performance based on analysis of the training data achieved an accuracy of 98%, with average weighted precision, recall and F1-score of 98%. Whereas, the testing evaluation yielded an accuracy of 99%, with an average weighted precision, recall and F1-score of 99%. The user satisfaction survey indicated that over 50% of the correspondents that participated in assessing the chatbot solution are highly satisfy with the performance of the system in correctness of grammar usage, contribution to a more efficient and time saving consultations, efficient use of preferred language, responses provided, prompt and correct response provisioning, and time saving in information acquisitions. The real-time testing of the chatbot was conducted locally on PC using the Visual Studio Code Editor running on a local server, and also through the internet with the assistance of NGROK.