Categorisation of sexual reproductive health short messages texts into thematic areas using text mining (a case study for the Zambia u-report).

In Zambia, the U-report platform was launched to improve young people's understanding of sexual reproductive health and access remote counseling services. This free texting program is run by the National HIV/AIDS/STI/TB Council supported by UNICEF. Millions of messages have been exchanged, but categorizing them manually to see what topics adolescents frequently ask about is challenging. This research explored using computers, to automatically classify these messages into different subject areas, making it faster to identify knowledge gaps in sexual health and related domains. The study investigated how U-report messages are currently categorized and built a system to automatically sort them into different topics. The new system was compared to the manual method currently used and it was found that the automated system is faster and more accurate, giving an accuracy score of above seventy percent. In the first stage of the research, administrators of the platform were interviewed to understand how they categorize messages. This helped identify the different categories they use. Then, a classification model was trained to sort messages into these categories using machine learning. This model has the potential to significantly improve how quickly and accurately messages are categorized on the Zambia Ureport SMS system.

Keywords—Feature Extraction, Machine Learning, Natural Language Processing (NLP), Text Classification, Text Mining, Text Extraction.