Short Video Illicit Promotional Content Detector on Chinese TikTok

Illicit promotional contents (IPCs) are advertising materials or campaigns that violate platform policies by employing deceptive, misleading, or fraudulent practices to promote products, services, or events. As short video platforms have gained popularity, these illicit contents have adapted to take the form of short videos,which we define as Short Video-Illicit promotional contents (SV-IPCs), becoming increasingly prevalent. Consequently, detecting SV-IPCs has become crucial in safeguarding users, especially minors, from fraudulent schemes and harmful contents. Existing detection schemes primarily focused on image processing, text analysis, and QR codes, making them ill-suited for short video platforms. To address this challenge, this paper presents the first comprehensive investigation into the SV-IPCs and their evasion techniques. By shedding light on the ecosystem underlying such illicit promotions, we propose a hybrid detection mechanism that combines natural language processing and video analysis techniques to deal with these emerging security challenges. Experimental results conducted on Chinese TikTok demonstrate the high effectiveness of the proposed scheme, achieving an impressive F1-score, recall, and precision of 90%, 90.9%, and 89.1%, respectively. Throughout the study, special attention is given to the potential implications of illicit promotion on short video platforms for both users and society as a whole.