Machine Learning in Ushahidi

Will Doran, Ushahidi’s senior developer, has an Msc in Machine Learning (ML), Data Modelling and Bioinformatics, so ML enhancements to the platform were inevitable.at the moment, Ushahidi has integrated two modules with the help of partner institutions.

The first is CREES (https://evhart.github.io/crees), a crisisevent classification system developed with the Knowledge Media Institute at the Open University. CREES analyses text and plucks out the crisis, information and event type. It then makes suggestions, for example, ‘I think these pieces of text relate to a fire’. This is really beneficial if you’re dealing with a large deployment, says Doran.

“In the Ushahidi platform you can pull from multiple data sources and get deluged with information. The idea of these tools is that they help people to divide the tasks and get to the actual impact state quicker.”

The other tool, YODIE (https://gate.ac.uk/applications/yodie. html), is an entity extraction tool which was developed with the Natural Language Processing Group at Sheffield University. Like

CREES, it scrutinises blocks of text but converts them from being two-dimensional to having semantic linking in a fashion similar to Google’s Knowledge Cards. “YODIE will pull out person, place and thing, and link them to Dbpedia,” explains Doran. Dbpedia is the open and free knowledge base that’s built from carefully extracting facts from Wikipedia and Wikidata. Both tools are part of the COMRADES project, a platform to support community resilience during a crisis.