Deep Learning Based Amharic Chatbot for FAQs in Universities
Project Overview
The document examines the implementation of a generative AI chatbot designed to support university students, particularly in engineering disciplines, by addressing frequently asked questions through deep learning and natural language processing techniques. This Amharic chatbot seeks to improve student engagement and knowledge acquisition by offering prompt, interactive responses. The study showcases the chatbot's impressive performance, achieving a 91.55% accuracy rate in delivering information, while also addressing the challenges associated with creating linguistic tools for under-resourced languages like Amharic. The findings suggest that such AI applications can significantly enhance educational experiences by making information more accessible, though they also highlight the need for further development in language processing capabilities for lesser-studied languages. Overall, the project illustrates the potential of generative AI to transform educational practices by fostering better communication and learning outcomes in diverse linguistic contexts.
Key Applications
Amharic chatbot for FAQs
Context: Higher education, specifically targeting engineering students at Mekelle University and Aksum University.
Implementation: Developed using deep learning techniques, including tokenization, normalization, and machine learning algorithms (SVM, MNB, DNN) for intent classification and integrated with Facebook Messenger and deployed on a Heroku server.
Outcomes: Achieved a 91.55% accuracy rate in responding to user queries, improved student access to information, and received a user satisfaction score of 86.2%.
Challenges: Challenges included the complexities of the Amharic language, lack of large datasets, and the need for grammatical tools for better response accuracy.
Implementation Barriers
Technical and Resource barrier
The Amharic language presents morphological richness and resource shortages, complicating chatbot development. There is a lack of comprehensive datasets for Amharic, limiting the chatbot's training and performance.
Proposed Solutions: Future research could explore the integration of Amharic WordNet and enhance the dataset by incorporating more Amharic FAQs. Recommendations include increasing the scope of research to gather more complex question types and expanding datasets for better training.
Project Team
Goitom Ybrah Hailu
Researcher
Hadush Hailu
Researcher
Shishay Welay
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Goitom Ybrah Hailu, Hadush Hailu, Shishay Welay
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai