9th Biennial ACSPRI Social Science Methodology Conference

Introduction
As social media usage continues to proliferate, platforms have become hotspots for recruitment into modern slavery. While research has predominately focused on recruitment into forced sexual exploitation, some efforts have been made to analyse recruitment into forced labour by utilising computational methods to gather and analyse online communications related to modern slavery activities (Williams, Burnap, & Sloan, 2017), albeit with limited attention to social media practices. Filling this gap is important to combating modern slavery as forced labour in the private economy makes up the largest share of global modern slavery (35%) and the majority of all forced labour cases (63%) (International Labour Organization, Walk Free & International Organization for Migration, 2022). Combating forced labour on social media is a significant challenge for stakeholders seeking to remove high-risk job advertisements, particularly those targeting prospective migrant workers, as traffickers embed content within visual and audio materials. This covert approach allows them to avoid detection by moderation protocols and obscures their criminal purpose from potential victims (de Vries & Radford, 2021).
This study bridges this gap by employing a multi-modal approach that integrates visual, auditory, and textual analysis. By examining social media data, this study aims to uncover and understand the complexity of how risks of modern slavery can present within recruitment and employment-related posts.

Methodology
Social media posts from six leading social media platforms (e.g., Tik Tok) were collected including textual, audio and visual data. These posts were related to job advertisements in the domestic work and construction sectors in the Middle East and North Africa, with a particular focus on countries within the Gulf Cooperation Council. Data preprocessing includes removing URLs and irrelevant characters to ensure uniformity. Audio data and video on-screen text were converted into text. Non-English posts were translated into English to maintain consistency in the analysis.
Natural Language Processing was employed to analyse textual data and detect risk markers of modern slavery, such as: 1) Topic modeling was applied to uncover prevalent themes related to exploitation and slavery within the collected data; 2) Pattern recognition algorithms were developed to identify specific linguistic patterns and keywords associated with potential modern slavery activities, such as references to coercion, restricted freedom, or abusive working conditions.; and 3) Co-word analysis was utilised as a visual technique to detect the co-occurrence of specific terms and concepts, facilitating the identification of clusters and patterns that can denote risk markers of modern slavery.

Conclusions
Our research has revealed several important findings. Findings show that potential traffickers’ social media posts format and structure tend to have more hashtags and limited captions, more text and descriptions within images or videos, and often contained multiple jobs in one advertisement post. We posit that these characteristics serve to appeal to as many prospective migrants as possible, using strategies to be viewed by large audiences on social media platforms, and sharing messages designed to entice and deceive. This research also highlights the challenges in analysing multi-modal data to determine whether ‘risk markers’ are present within social media posts where there is limited contextual data based on the sector, country, or language.
This research significantly advances the fields of digital criminology and social media analytics by developing a novel multi-modal analytics approach to uncover and understand the complexity of modern slavery risks within recruitment social media posts. The findings can aid law enforcement agencies, NGOs, and social media platforms in the detection and mitigation of modern slavery risks on social media, contributing to global efforts in combating this critical issue.