Enterprise AI Analysis
Deep Learning-based Amharic Chatbot for FAQs in Universities
This paper presents a deep learning-based Amharic chatbot designed to assist university students with frequently asked questions (FAQs). Leveraging natural language processing (NLP) and deep neural networks, the chatbot achieves 91.55% accuracy in classifying Amharic queries. It addresses challenges like morphological variation and lexical gaps, providing instant information retrieval. Deployed on Facebook Messenger via Heroku, the system enhances user experience and reduces administrative burden.
Key Impact Metrics
0
Accuracy Rate
0
F1-Score
0
Validation Loss
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Problem Background
University students often struggle to find answers to common questions, leading to wasted time for both students and administrative staff. Current solutions are often slow and ineffective. Chatbots offer a promising solution for immediate, up-to-date information, but developing them for morphologically rich languages like Amharic presents unique challenges due to limited resources and complex linguistic structures.
Methodology
The research followed a Design Science Research (DSR) approach, encompassing problem identification, objective definition, design, development, demonstration, and evaluation. Data collection involved questionnaires from engineering students at Mekelle and Aksum Universities, translated and preprocessed into a JSON dataset of 60 topics.
Chatbot Framework
The chatbot framework consists of user, user interface (Facebook Messenger via Flask webhook), and the chatbot model. The model has three parts: intent classification (preprocessing, tokenization, normalization, stop word removal, stemming, Bag of Words, classification), training (with DNN using TensorFlow/Keras/NLTK), and response generation (selecting appropriate response based on identified intent).
Experiments & Results
The study compared SVM, Multinomial Naïve Bayes, and Deep Neural Networks (DNN) for intent classification. DNN, with ReLU activation and two hidden layers (128 and 64 neurons), achieved the best performance: 91.55% accuracy, 85.98% precision, 87.13% recall, and 85.23% F1-score. The model was trained for 150 epochs with an Adam optimizer.
Evaluation & Contribution
User satisfaction (86.2%) was measured using a Likert scale across dimensions like grammar restriction, robustness, Amharic alphabet variation, ease of use, user friendliness, and answer validity. The chatbot significantly contributes to NLP and Amharic language technology as the first Amharic FAQ chatbot for universities.
91.55%
Deep Learning Achieves High Accuracy in Amharic Intent Classification
Enterprise Process Flow
Input Text
→
Pre-processing (Tokenization, Normalization, Stop-word Removal, Stemming)
→
Bag of Words (BOW)
→
DNN Classifier
→
Intent Classification
→
Response Generation
| Classifier | Accuracy | Precision | Recall | F1-Score |
|---|---|---|---|---|
| DNN (ReLU) | 91.55% | 85.98% | 87.13% | 85.23% |
| SVM | 87.96% | 80.9% | 86.0% | 78.4% |
| Multinomial Naïve Bayes | 60.19% | 50.5% | 54.7% | 49.6% |
Amharic Chatbot Deployment
The developed Amharic chatbot was successfully integrated with Facebook Messenger and deployed on a Heroku server, ensuring 24-hour accessibility for university students. This deployment validates the system's practical applicability in real-world scenarios, offering instant support and reducing the burden on administrative staff. The use of familiar platforms like Messenger boosts user adoption and engagement.
Outcome: Improved student support and administrative efficiency.
The study addresses significant challenges in Amharic NLP, including morphological richness, Fidel variation, and lexical gaps. By leveraging deep learning techniques, the chatbot demonstrates a robust approach to handling these complexities, paving the way for more advanced Amharic conversational AI systems. Future work aims to integrate sentiment analysis and expand language support to other Ethiopian languages.
Calculate Your Potential AI ROI
Estimate the efficiency gains and cost savings your enterprise could achieve by implementing AI solutions based on this research.
Potential Annual Savings
$0
Annual Hours Reclaimed
0
AI Implementation Roadmap
A phased approach to integrate deep learning-based chatbots into your enterprise, maximizing success and minimizing disruption.
Phase 1: Data Collection & Preprocessing
Gathering and cleaning Amharic FAQ data, including translation and normalization.
Phase 2: Model Training & Optimization
Training Deep Neural Network models using TensorFlow/Keras and optimizing for accuracy.
Phase 3: Integration & Deployment
Integrating the chatbot with Facebook Messenger and deploying on a cloud server for accessibility.
Phase 4: Evaluation & Refinement
Conducting user satisfaction surveys and iterative model improvements based on feedback.
Ready to Transform Your Enterprise with AI?
Book a personalized consultation to explore how these advanced AI insights can be tailored to your specific business needs and drive tangible results.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat