DEVELOPING AFAAN OROMO CHATBOT FOR HIV/AIDS PREVENTION AND CARE COUNSELING USING DEEP LEARNING APPROACHES

Abstract

Chatbots or conversational chatting machines are being built using Artificial Intelligence and machine learning approaches to solve the existing problems in the area of natural language processing. Advanced intelligent conversational chatbot systems have been developed as a result of recent developments in the human-computer interaction field, where text-based chatbot frameworks are a popular focus of research for both industry and academia. Recently, the use of chatbots has grown rapidly in various fields such as marketing, assistance systems, education, healthcare, cultural heritage, entertainment, etc. In this study, we design and implement deep learning-based chatbots for HIV/AIDS using the Afaan Oromo language for users. For this study, a dataset with question-and-answer pair statements has been collected by filtering the frequently asked questions from the different healthcare organization websites and national comprehensive HIV prevention, care, and treatment guidelines and prepared in JSON format. The prepared dataset has been passed through the appropriate data preparation before being used for model training purposes and the word2vec word embedding method was used as a feature extraction technique. For the model design purpose, convolutional neural network (CNN), long short-term memory (LSTM), bidirectional long short-term memory (BiLSTM), and bidirectional gated recurrent unit (BiGRU) were applied and 92.11%, 93.8%, 95.27%, and 94.4% accuracy achieved respectively. The stratified k-fold cross-validation has been applied to each model during the training session to hold overfitting challenges. Finally, human evaluation is applied to the system's performance and system acceptability. The average result of system performance has achieved 80.5% and the user acceptance testing average result has achieved 84.12% which shows promising results in acceptance and performance of the system