Accelerating UN Humanitarian Aid Planning

Case Study

“If we can see clearly what the people’s priorities are, and where they are, this creates a huge opportunity to expand our analysis.”
- Jonathan Brooker, OCHA ROMENA Team

The challenge of finding data that's fit-for-purpose isn't unique to OCHA. Humanitarian response teams require a thorough knowledge of what's happening on the ground in disaster areas to develop and implement an effective response plan. Additional qualitative and quantitative insight and analysis are required to understand the circumstances that drive a need for humanitarian aid in regions too dangerous for aid workers to visit.

OCHA's Libya country team kept track of approximately 330 distinct data sources daily -- including Twitter and Facebook, local radio stations, TV channels, reports from contacts on the ground, and news publications -- to track what people in monitored regions were talking about from day to day. Manually gathering this data is time-consuming, and the data's meaning was often ambiguous. As a result, humanitarian aid planners are often forced to do with scant and inadequate information, and response teams must build to leverage data that is often incomplete and anecdotal.

Our team initiated this project in partnership with OCHA's office for the Region of the Middle East and North Africa (ROMENA), Microsoft, and UN Global Pulse to examine the potential of analysis of publicly available data to help OCHA ROMENA improve their ability to monitor, assess, and respond to humanitarian needs in Libya.

Identifying a Partner

Before we were even aware of this particular challenge or the UN office we'd be working with, our team's first step was to solicit UN offices for problem areas we could help support. We received challenges from offices including the Department of Economic and Social Affairs, Political and Peacebuilding, Peace Operations, and Global Communications. After conducting interviews with departments, we identified OCHA's challenge as the best fit for our collaboration teams to have the most impact.

Discovery

We began by working with the OCHA Libya team to understand their current experience. They joined us in New York for rounds of interviews, current and future state journey mapping, and embedded throughout early prototyping and testing.

The final map described key stages in the Libya team's information management cycle, across information collection, consolidation, cleaning, analysis, and ultimately making response decisions. The map helped us understand important challenges at each step, opportunities for improvement and ideal scenarios, and the current toolset the team used along each step.

The Libya team facilitated a fact-finding mission in Tunis, helping us get a much clearer view of the day-to-day work on the ground.

From this work, we began to develop a database of their go-to information sources, including news and regional media outlets, TV stations, Facebook pages, Twitter accounts, blogs, and radio stations. This list would become the seeds of our eventual data pipeline.

From there, we worked together to develop a massive list of keywords, in Libyan Arabic and English, that the team uses when they're searching for actionable information.

Prototyping

Through prototyping, we zeroed in on our approach: to build, test, and improve a map-based dashboard that provides the Libya team with data-driven insight into the current needs for humanitarian aid and helps create more accurate and practical plans to respond to these challenges. Our combined teams laid the conceptual and technical groundwork during a design and dev sprint in New York at Microsoft headquarters.

We determined the dashboard would analyze and plot data that met specific criteria (for example, the inclusion of a large or related set of keywords) on a map using location data, visual indication of sentiment analysis, and aggregate analysis of data in geographic or topical clusters.

Microsoft developer Erik Schlegel explains:

"At a high level, we designed and built a data ingestion, analysis, and visualization pipeline. The pipeline collects social media conversations and postings from the public web and darknet data sources. It then performs feature extraction and infers relationships between targeted keywords in real-time.
‍
Conversational message streams are paired with sentiment analysis and mood inference modeling alongside other machine-learning techniques to gain quantitative insight into the topics, demographics, and indicators driving vital humanitarian conditions for a targeted set of locations. 
‍
Finally, results are visualized on a dashboard so that a user can see trending topics and keywords over time and geography."

The early prototype used a live data pipeline to populate the dashboard. In the image below you can see clusters of conversation based on keywords color-coded by sentiment where green is positive, yellow is neutral, and red is negative.

‍Piloting

The project moved into the pilot stage. During this stage, the Libya team ran the technology in parallel with its standard practices, continuing to train the model and report back on its accuracy. This research guided and informed this project's ability to scale and inform the team about future public information and media monitoring technologies opportunities.

The result became Project Fortis, which was used by the OCHA ROMENA team and other regional UN offices. 

Incoming data is filtered, analyzed with Natural Language Processing, and run through a machine learning service learning to infer location and conversations related to key humanitarian clusters - including Early Recovery, Food Security, Health, Protection, and Water, Sanitation, and Hygiene.

Watch a demo of the dashboard by OCHA Columbia below. (The video is in Spanish. If you don't speak Spanish, select the English-translated subtitles.

‍
The technology itself was released under an open-source, MIT license. A long-term objective for this project is to create a shareable resource for other humanitarian groups and UN agencies that can be adapted, learned from, and reused across a variety of regional contexts and related challenge areas.

An example use came from Sweden's Umeå University, which used the tool to forecast areas at high risk of dengue fever in Sri Lanka and Indonesia.

Recognition

The project was selected for the Humanitarian Innovation Showcase at the 2016 World Humanitarian Summit in Istanbul. We had the opportunity to share the project with Secretary General Ban Ki-moon as he visited our booth in the showcase.

Github

You can view, download, and contribute to the project on Github here.