The goal of this two-month Omdena AI Challenge was to build an MVP for a monitoring tool for human rights abuses. The challenge partner Human Rights First is a non-profit, non-partisan international human rights organization with a track record of success in delivering change that has made a meaningful difference in people’s lives. Whether protecting refugees, combating torture, or defending persecuted minorities, they focus not on making a point, but on making a difference; for over 30 years.

The problem

With the evolution of social media in the last decade, the world’s ability to communicate has increased exponentially. The abundance of data this communication generates provides an opportunity to evaluate social phenomena across the globe in near real time. In particular, it can enable the identification of reported human rights abuses almost immediately. Timely identification can allow for timely aid or action from governments and civil society to prevent or reduce harm.

Human Rights First about the project outcomes

The project outcomes

The final Dashboard MVP

The proposed solution is a real-time monitoring system that aims to collect and analyze data from major social media platforms such as Twitter, Reddit, Instagram, and Telegram. By utilizing a set of ensemble language models, the application can identify text that is likely to represent human rights abuses.

For instance, advanced BERT-based models can cluster, label, and compare text to identify “ground truth” instances of “human rights abuses” in written language. The goal is to use the system as an early warning mechanism that identifies, characterizes, and visualizes human rights abuses on a front-end dashboard.

Characterization involves identifying language that is indicative of human rights abuses and categorizing them into clearly defined classes based on qualitative foundations. The aim of this effort is to develop an application with a back-end capable of collecting, storing, categorizing, and visualizing social media data at scale in real-time.

The analytics involved will extract social media posts that describe human rights abuses, along with associated confidence intervals and a severity score based on the frequency of unique incidents.