Blog

Using big data and machine learning to respond to the refugee crisis in Uganda

Paula Hidalgo-Sanchis, Manager, Pulse Lab Kampala
Mar 5, 2018

Applying AI to transform public radio content into insights for development and humanitarian action 

Hosting over 1.2 million refugees from South Sudan, Uganda has one of the most favorable refugee protection environments in the world, providing refugees with freedom of movement, the right to work, and access to social services through a generous asylum policy (the Refugee Act of 2006 and the Refugee Regulations of 2010). “ […] Uganda embraces children […] Uganda allows us to go to school,” says Hindiyo Abdulkadir, a young representative of refugees speaking at the Uganda Solidarity Summit on Refugees.

The country is experiencing the fastest growing refugee community in the world that has a far-reaching economic and social impact. Most settlements for South Sudanese refugees are based in the north of Uganda where an average of 2,000 refugees arrive every day. The mass influx to this poor area of the country is putting a strain on resources like land, firewood and water. In this context, tensions have already emerged between the refugees and the hosting communities and the shortages in funding represent a growing challenge that the Government, the UN, and humanitarian actors must address.

UN Global Pulse, through its lab in Kampala, has been working with the UN and the Government to explore how insights from public talk radio can provide real time information on what is happening on the ground. The hypothesis is that the systematic analysis of what people say on the radio regarding their situation, concerns, and needs provides actionable insights for programme implementation.

In the private sector, the analysis of public discussions on social media is regularly used for marketing, advertising and management. The same type of analysis can be used by development and humanitarian practitioners to complement traditional means for gauging public sentiment (such as surveys or focus groups discussions). This type of analysis can be cost-efficient and provide real time insights to feed early warning systems.

However, in Uganda, social media platforms are still in incipient phase -- for example, Facebook subscribers amount to only 2.2 million (roughly 5% of the population). Radio on the other hand is the main source of information, reportedly used by 55% of the households in the country. Calling into a talk show on radio to share thoughts, opinions and reports is a common practice especially in rural areas.

Pulse Lab Kampala, supported by the United Nations Development Operations Coordination Office (UNDOCO), has developed a technology prototype that allows analysing public discourse on radio the same way that it is analysed on social media. The Lab and its partners used artificial intelligence and machine learning to develop a digital application called the Radio Content Analysis Tool. The tool transforms discussions taking place on radio into big data text, which is then mined for topics of interested relevant to the Sustainable Development Goals (SDGs).

The technology provides a flow of real-time information that development and humanitarian agencies can use to understand what people think and how they feel, and to channel the country’s collective intelligence to gain insights and solve problems. The usability of the technology was tested with a series of pilot studies that were summarized in the report “Using machine learning to analyse radio content in Uganda – Opportunities for sustainable development and humanitarian action.”As part of the pilot studies, Pulse Lab Kampala and its partners also conducted a large-scale exploration to support the response of the UN Country Team and the Government to the refugee crisis in Uganda.

The analysis was conducted from July 2016 to February 2017. Filters to detect conversations regarding refugees were created and a translation team annotated all mentions of refugees, including issues of acceptance, health or increasing social tensions. To be able to identify the level of concern of host communities towards refugees, the project conducted an inductive and manual tagging of relevant keywords.

The study demonstrated that speech-to-text processing (by automated dictation software), combined with natural language processing (with software that can interpret text), may be used to close an information gap – it can provide insights on early warning mechanisms, can help to monitor the implementation of projects and programmes, and can support programme evaluation -- by generating machine-readable big data about refugees in rural areas where up-to-date information of any kind is seldom available.

1. Radio testimonials and reports can inform early warning systems. In an early warning system, qualitative information is as valuable as quantitative information as the ultimate goal is to provide timely alerts to provoke action such as aid delivery programmes.

 “The reduction of food rations in refugee settlements in West Nile, is leading to increased cases of domestic violence” explains the conductor at Arua 90.9 FM radio station (November 2017). A young refugee from Ofua III block narrates then how the food ration has been reduced from 12 to 6 Kgs with a cash allowance of 7,000 Ugandan Shillings to make up for the shortfall.

2. Opinions and rumours recounted on radio can help with monitoring evolving contexts. Insights on perceptions, misconceptions or hate speech can provide valuable information not only to monitor but also to course-correct programmes and operations in real time. For example, in Uganda many attempted health interventions are resisted and hindered due to circulating rumours and misconceptions. To address this, it is important that healthcare delivery programmes understand these rumours and the context in which they spread.

“…I was actually tempted to think that the source of the anthrax outbreak in West Nile was due to the animals that entered our country (to refugee settlements) without being screened…”.

3. Reports shared by citizens on radio can support programme evaluation. Big data can often provide real-time and continuous data, which is helpful for observing the processes through which a programme evolves ( report - Integrating Big Data into the Monitoring and Evaluation of Development Programmes). 

“…there were donations to help with buying seedlings for planting to assist the refugees with food…only a few people were given these seedlings and the rest of the seeds were used in personal farms of these government officials…”.

The analysis showed that public radio data can enable more agile, efficient and evidence-based decision-making and can better measure progress on the SDGs. There now exists an unprecedented opportunity to invest in data-driven innovations that save lives through smarter decision-making and more efficient ways of working. The opportunity also exists to mobilize public sector, private sector and civil society to work together to overcome persistent barriers to research, data sharing, scaling and responsible use of data for development and humanitarian action.