A Multi-Input Machine Learning Approach to Classifying Sex Trafficking from Online Escort Advertisements

Sex trafficking victims are often advertised through online escort sites. These ads can be publicly accessed, but law enforcement lacks the resources to comb hundreds of identify those that may feature sex-trafficked individuals. The purpose this study was implement and test multi-input, deep learning (DL) binary classification models predict probability an ad being associated with sex (ST) activity aid in detection investigation ST. Data from 12,350 scraped classified were split into training sets (80% 20%, respectively). Multi-input included recurrent neural networks (RNN) for text classification, convolutional (CNN, specifically EfficientNetB6 or ENET) image/emoji (NN) trained used classify 20% set. best-performing DL model imagery inputs, resulting accuracy 0.82 F1 score 0.70. More importantly, best classifier (RNN + correctly identified 14 sites had estimates 0.845 greater (1.0 precision); precision 96% multi-input (NN RNN when only highest positive probabilities (>0.90) considered (n = 202 ads). developed could productionalized piloted criminal investigators, as they potentially increase their efficiency identifying potential ST victims.