Domestic Violence Crisis Identification From Facebook Posts Based on Deep Learning

Abstract

Domestic violence (DV) is a cause of concern due to the threat it poses toward public health and human rights. There is a need for quick identification of the victims of this condition so that DV crisis service (DVCS) can offer necessary support in a timely manner. The availability of social media has allowed DV victims to share their stories and receive support from the community, which opens an opportunity for DVCS to actively approach and support DV victims. However, it is time consuming and inefficient to manually browse through a massive number of available posts. This paper adopts deep learning as an approach for automatic identification of DV victims in critical need. Empirical evidence on a ground truth data set has achieved an accuracy of up to 94%, which outperforms traditional machine-learning techniques. The analysis of informative features helps to identify important words which might indicate critical posts in the classification process. The experimental results are helpful to researchers and practitioners in developing techniques for identifying and supporting DV victims.