Developing amharic text-based chatbot model for HIV/AIDS awareness and care using deep learning approaches

The healthcare sector relies heavily on human interaction, yet conversational AI applications like chatbots are becoming increasingly important. While global standard HIV/AIDS patient assistance conversational AI bots exist, they often lack compatibility with diverse languages and the specific contexts of users' living standards in various countries. This study proposes an Amharic text-based chatbot model designed for HIV/AIDS awareness and care advising, utilizing deep learning algorithms. We collected and prepared 10,291 text data entries from reliable sources covering common HIV/AIDS topics, converting them to a semi-structured JSON format. After preprocessing through cleaning, normalization, tokenization, and embedding, FastText was used for feature extraction. We evaluated different deep learning algorithms, including LSTM, BiLSTM, GRU, and BiGRU, using accuracy, precision, recall, F1-score, and human evaluation. Extensive experiments were conducted, and the Adaptive Moment Estimation (Adam) optimizer outperformed the Stochastic Gradient Descent (SGD) optimizer. To prevent overfitting, dropout and early stopping techniques were employed. The BiGRU model achieved the highest performance, with a testing accuracy of 95.01% and a loss of 0.372, based on an 80:20 train-test split, a batch size of 32, 50 epochs, and a learning rate of 0.0001. The developed Amharic text-based chatbot model demonstrates promising performance in providing HIV/AIDS awareness and care advice, with the BiGRU model proving to be the most effective algorithm for this task. This model has the potential to improve access to vital health information for Amharic speakers, addressing linguistic and contextual barriers present in existing global solutions.