Afaan Oromo Text-based Consultant Chatbot for Maternal Healthcare Using Deep Learning

Abstract

Healthcare is very essential to leading a good life. In healthcare fields, maternal healthcare is a health service provided to pregnant women to retain their health safely. It is one of the essential strategies in decreasing maternal mortality and is provided by maternal health care providers. They offer a better understanding of pregnancy-related health problems. Every mother wants to get relevant advice every day, from prenatal up to postpartum. In developing countries, most expectant mothers do not get adequate prenatal care from health care providers. Because the ratio of demand and maternal health experts do not match; the health center faces a range of challenges to address the service for all users timely. Due to this reason, maternal and infant mortality remained high in developing countries. Currently, Ethiopia is identified with the high maternal mortality among other developing countries in the world due to poor maternal healthcare provision. Nowadays, technology has introduced the techniques of getting access to information simply and efficiently. One example of such technology is a chatbot model that can be considered the best way to assist a user in all industries. Because the chatbot can imitate humans and is available 24/7 hours to respond quickly to questions users ask. A chatbot solution can offer better information for pregnant women on known danger signs identification, nutritional tips, prenatal tips, family planning, noticeable symptoms, etc irrespective of time and place. This study introduced the Afaan Oromo text-based maternal health consultant chatbot using deep learning to tackle the challenges formerly mentioned. To accomplish the goal of this study, the dataset was collected from ORHB, online websites (https://fayyaa.net/page/1/), and some hospitals in Oromia to build the maternal health consultant chatbot model. The collected dataset was prepared and preprocessed using various NLP techniques. And then, the Word2Vec model is utilized to represent text data with vectors. In this work, CNN, CNN-LSTM, BiLSTM, and CNN BiLSTM are presented and evaluated in terms of their performance to design a maternal health consultant chatbot. Based on the experiment performed, the CNN-BiLSTM model outperformed the other proposed models with 96.94% accuracy under Stratified 5-fold cross validation. This study improves the performance of the existing chatbot model by merging the CNN model with the BiLSTM model to design a maternal health consultant chatbot model.