Deep Learning Based Amharic Chatbot for FAQs in Universities
Summary: arXiv:2402.01720v3 Announce Type: replace-cross
Introduction
University students often face challenges in obtaining timely answers to frequently asked questions from administrators or faculty members. This process can be time-consuming and frustrating for both students and university staff. To address this issue, researchers have proposed a novel solution—a chatbot model specifically designed to facilitate communication and provide accurate responses to common queries in the Amharic language.
Chatbot Overview
The proposed chatbot leverages natural language processing (NLP) and deep learning techniques to enhance the interaction between students and university personnel. Chatbots are essentially computer programs that utilize artificial intelligence (AI) to simulate human conversation, functioning as virtual assistants to manage inquiries and perform various tasks.
Methodology
The chatbot utilizes several key techniques to effectively analyze and respond to Amharic input sentences:
- Tokenization: The process of breaking down sentences into individual words or phrases.
- Normalization: Adjusting the text to a standard format to improve processing accuracy.
- Stop Word Removal: Eliminating common words that do not contribute significant meaning to the queries.
- Stemming: Reducing words to their root form to enhance the understanding of queries.
Machine Learning Algorithms Used
To classify tokens and retrieve the most relevant responses, the chatbot employs three distinct machine learning model algorithms:
- Support Vector Machine (SVM): A powerful classification technique that aims to find the optimal hyperplane separating different classes.
- Multinomial Naïve Bayes: A probabilistic model based on Bayes’ theorem, particularly effective for text classification tasks.
- Deep Neural Networks: Implemented using TensorFlow and Keras, this approach has shown superior performance in complex data scenarios.
Results
The deep learning model outperformed other algorithms, achieving an impressive accuracy rate of 91.55%. Additionally, it recorded a validation loss of 0.3548, utilizing the Adam optimizer and SoftMax activation function to enhance its learning capabilities.
Deployment and Integration
The chatbot was successfully integrated with Facebook Messenger, providing users with easy access to its functionalities. Furthermore, it was deployed on a Heroku server, ensuring 24-hour availability for students seeking answers to their queries.
Challenges and Future Research
The experimental results indicate that the chatbot framework effectively met its objectives, successfully addressing challenges such as:
- Amharic Fidel variation
- Morphological variation
- Lexical gaps in the language
Looking ahead, future research could focus on the integration of Amharic WordNet to further narrow the lexical gaps and enable the chatbot to handle more complex questions, ultimately improving the user experience and enhancing educational communication within universities.