The third speaker in this ACSPRI 2024 conference session is Mingming Cheng, whose focus is on the use of multi-modal data in the analysis of modern slavery risks on social media. Modern slavery includes forced labour, forced marriage, human trafficking, debt bondage, and other related practices; it targets vulnerable individuals, and these might also be identified through social media (for instance through ads or recruitment posts that mimic ordinary advertising).
Such content can spread very rapidly, and identifying and regulating it is very difficult. It requires a multi-modal approach that analyses visual, audio, and text data, and the present project does so for six social media platforms (TikTok, LinkedIN, Facebook, Twitter, YouTube, and Instagram), using a range of terms for ‘housemaid’ in relevant languages. This identified some 575,000 matching posts, and these were then processed for their multi-modal content.
Offers of free aspects (e.g. ‘free visa’) and a sense of urgency were common in such posts; jobs and job locations were described very generically, and some conditions appeared too good to be true; multiple hashtags covering all possible keywords were also very prominent in the posts. Posts offered multiple job opportunities, circumvented content moderation by embedding content in videos, and pointed interested audiences to third-party messaging platforms (WhatsApp, Snapchat) with considerably less public scrutiny.
Visual content embedded in videos makes up nearly half of the relevant content here; this creates considerable analytical challenges. Content is also in a range of languages, targeting specific populations. There is also considerable use of jargon and coded language which may not be easily understandable by outside researchers.
