Detection of Possible Illicit Messages Using Natural Language Processing and Computer Vision on Twitter and Linked Websites

Abstract

There is a worldwide epidemic of human trafficking that diminishes the self-respect of millions of victims. When it comes to promoting illegal services on social media, covert communications are being utilized to spread the word. As resources for law enforcement are limited, it is critical that messages that may be connected to this crime and serve as clues be automatically detected. In this paper, we use natural language processing to identify Twitter tweets that potentially promote illegal services and exploit kids. It is now feasible to identify photos of children as young as 14 years old among the images and URLs identified in suspicious messages. We followed this procedure to conduct our research. The first step is to mine real-time tweets using hashtags relating to minors. After the tweets have been cleaned out of extraneous noise and typos, they are either labeled as suspicious or not. Haar models are also used to select facial and torso geometric features. We can recognize gender and age group by using Support Vector Machine (SVM) and Convolutional Neural Network (CNN), even when the facial details are blurred, by using torso information and its proportional relationship with the head. Because of this, the SVM model with just torso features outperforms CNN.