Minority report detection in refugee-authored community-driven journalism using RBMs

TL;DR

This paper introduces a novel approach combining stakeholder data collection, deep Boltzmann machine-based topic modeling, and anomaly detection to identify minority reports in refugee community journalism, aiding policy and human rights efforts.

Contribution

It presents a new methodology integrating deep learning and anomaly detection for analyzing refugee narratives to uncover minority reports and human rights issues.

Findings

Effective identification of minority reports using RBMs

Enhanced mapping of human rights violations

Improved detection of sensitive issues in refugee stories

Abstract

Our work seeks to gather and distribute sensitive information from refugee settlements to stakeholders to help shape policy and help guide action networks. In this paper, we propose the following 1) a method of data collection through stakeholder organizations experienced in working with displaced and refugee communities, 2) a method of topic modeling based on Deep Boltzmann Machines that identifies topics and issues of interest within the population, to help enable mapping of human rights violations, and 3) a secondary analysis component that will use the probability of fit to isolate minority reports within these stories using anomaly detection techniques.