Skip to content

Improving health chat helplines in South Africa

How we are using natural language processing to strengthen health information sharing

Photo credit: Angelo Moleele on Unsplash

Decision-maker’s challenge

The South African government runs WhatsApp-based public health helplines for its citizens and its technology partner saw an opportunity to improve this service delivery. Human experts are needed to spend substantial time on individual cases before providing helpful answers to citizens, especially when it comes to complicated health areas like sexual health, COVID-19, pregnancy, etc. Users often have to wait long hours before receiving useful advice.

Impact opportunity

The maternal health helpline serves more than four million women, most of whom send their questions directly to a team of help-desk operators. But a user base this big can easily overwhelm even the most dedicated team; in peak times, the team of three receives more than 1200 questions per day. And in the context of maternal health, where the situation could be life-threatening, it is very important for them to quickly and accurately answer a high volume of incoming questions.

The technology partner needed an automated way to respond to some questions with an FAQ best answering them so that mothers could get a rapid resolution for common concerns and the help desk could focus on nuanced and urgent cases that required high-level expertise.

Our approach

IDinsight built a plug-in called “Ask-A-Question (AAQ)” powered by Natural Language Processing (NLP) technology that automatically forwards pre-saved health content best addressing a user’s question. With an ability to correct misspelt words, process grammatically incorrect sentences and comprehend colloquial terms, AAQ stands at the forefront of modern technology deployed at a national scale.

Apart from the core FAQ-matching algorithm in AAQ, the IDinsight team delivered a number of complementary features to make AAQ easier to work with, including an admin web application, an urgency detection module and a monitoring dashboard.

The results

The latest pilot deployment of AAQ has a top-five accuracy of ~65 percent when assessed on actual historical chat data, i.e. the correct FAQ appears in the top five FAQ suggestions 65 percent of the time. 

We are continuously upgrading the core AAQ algorithm and will be incorporating real-user feedback to make further improvements. AAQ will eventually be open-source, so it can be used and contributed to by many other help-desk organizations in the social sector.

Relevant resources: