Over the past year, the Umati project team has been building on tools created over 2014 to automate the Umati monitoring project. This is in line with one of the project's key objectives in the second phase with an aim to increase efficiency and scalability of the project.
READ MORE ON THE UMATI PROJECT HERE.
We are pleased to share an update report on the progress of automating the Umati Project. We have been successful in automating data collection, and initial filtering out of noise, and continue to automate aspects of analysis as well. However, we are cognisant of the fact that human input is still invaluable; the aim of introducing automation is to augment the efforts of humans in the process - context, for instance is not something a machine can learn.
This report explains the project's background and methodology as applied in phase 1 that was human-intensive, and goes on to describe how various aspects of the monitoring process are primarily conducted by algorithms, as well as where human input comes in. This 'intelligent monitor' is envisioned as a unified, open-source tool that can be used for online hate/dangerous speech monitoring not only in Kenya, but in other country contexts, where the subject is gaining interest. The project is currently being replicated in Nigeria and South Sudan.
We will also devise a toolkit with step by step considerations for replicating Umati, and we intend that this will receive input from other deployments of online speech monitoring around the world.
Feedback is most welcome, and we invite partners to collaborate with us in building the Intelligent Umati Monitor or analysing the Umati process. If interested, please get in touch on umati[at]ihub[dot]co[ke].
You can access the Umati Intelligent Monitor report here.
Previous blog posts on the Umati project can be accessed here.