Innovating with Generative AI to Better Understand Community Needs
Image credit: UNHCR
The United Nations High Commissioner for Refugees (UNHCR) Innovation Service in El Salvador has developed a novel approach to community needs assessment using a secure generative AI tool, dubbed SIVAR+. This tool leverages a large language model, trained exclusively on a dataset of focus group discussion (FGD) transcriptions, to provide concise summaries of community feedback on specific topics.
The development process involved four stages: gathering high-quality audio recordings, converting speech to text, cleaning and anonymizing the transcriptions, and mining the data for insights. The team utilized open-source tools, including Sonix and Whisper, to achieve an accuracy rate of approximately 80%. To mitigate the risk of "hallucinations," the tool was designed to only utilize data from the transcripts.
The SIVAR+ tool enables users to interact with the data in a natural language interface, providing broad information on expressed needs and specific details upon request. This innovation has significantly reduced the time required to compile planning and results reports, from two to three months to less than half that time. The tool has also informed the ongoing Azure OpenAI collaboration between the Innovation Service and UNHCR's Division of Information Systems and Telecommunications.
Furthermore, the project aimed to build coding capacity in the community by providing an online course in R programming to 23 community members at risk of displacement. The participants received a Google-backed diploma upon completion and demonstrated a practical understanding of data analysis skills.
The successful implementation of SIVAR+ has the potential to revolutionize community needs assessment and inform decision-making processes within the UNHCR. The project's learnings will contribute to the responsible and creative leveraging of generative AI and other AI technologies to further the mission of UNHCR.
