Abstract
The Philippines has emerged as a global center for child cybersex trafficking, with perpetrators increasingly leveraging social media platforms like Twitter to distribute illicit content. This study employs Natural Language Processing (NLP) techniques, including word co-occurrence networks and short-text clustering (GSDMM), to analyze the linguistic patterns and operational roles of these traffickers. By examining a dataset of nearly 33,000 unique tweets, the research identifies a highly structured lexicon and categorizes malicious actors into four distinct personas: The Propagator, The Peddler, The Social, and The Voyeur. These findings provide a framework for law enforcement to better detect and mitigate online child exploitation.
Introduction
The Philippines was identified in a 2018 UNICEF survey as a primary host for child sexual abuse materials (CSAM). Traffickers utilize anonymous "alter" (alternate) accounts on Twitter to peddle and trade illegal content, often evading detection through coded language and colloquialisms. Understanding the specific writing styles and structural roles within these networks is essential for effective cybercrime intervention. This paper addresses this gap by mapping the linguistic landscape of child cybersex trafficking in the Philippines.
Methodology
The study treats the identification of trafficker behavior as a text analysis and clustering problem, utilizing three primary techniques:
- Word Cloud Visualization: To identify high-frequency colloquialisms and explicit terms.
- Trigram Co-occurrence Networks: To map sequences of three words that frequently appear together, uncovering structural patterns such as calls to action.
- Gibbs Sampling Dirichlet Multinomial Mixture (GSDMM): A short-text clustering algorithm optimized for the brevity of tweets. GSDMM aggregates vocabulary into distinct clusters, which are then mapped to specific user "personas."
Dataset and Filtering
The analysis utilized 69,675 tweets collected between October 2019 and July 2020, restricted to the Philippines via geographic bounding boxes. The data was filtered using specific hashtags known to be used by traffickers, such as #bagets (slang for children) and #sarapngbagets. After deduplication and removal of retweets, a final corpus of 32,899 unique tweets was analyzed.
Results and Discussion
Structural Linguistic Patterns
The trigram co-occurrence network reveals that traffickers employ a repetitive, structured vocabulary to manipulate platform algorithms and drive engagement. The network formed three distinct subgraphs representing different operational activities: crediting/trading (e.g., "ctto", "trade"), proliferation (e.g., "rt", "follow", "dm me"), and content organization (e.g., "open thread").
Operational Personas
Using GSDMM clustering, the study identified four distinct operational roles based on vocabulary usage. This division of labor suggests a sophisticated ecosystem where different accounts handle marketing, financial transactions, and content acquisition.
- The Propagator: Focuses on content reach through retweets and follows.
- The Peddler: Manages the illicit marketplace, handling pricing and availability.
- The Social: Encourages meetups or direct interactions.
- The Voyeur: Targets minors for voyeuristic content, often using hidden cameras.
Conclusion
This investigation highlights the structured nature of child cybersex trafficking on Twitter. The identification of specific personas and their associated lexicons provides a foundation for operationalizing these findings. Future work should involve partnering with child protection agencies to implement these NLP models for real-time tracking and apprehension of traffickers.
Summary updated