Degree

Doctor of Business Administration (DBA)

Department

The Stephenson Department of Entrepreneurship and Information Systems (SDEIS)

Document Type

Dissertation

Abstract

Human trafficking or modern slavery is a problem that has plagued every U.S. state, in both urban and rural areas. During the past decades, online advertisements for sex trafficking have rapidly increased in numbers. The advancement of the Internet and smart phones have made it easier for sex traffickers to contact and recruit their victims and advertise and sell them online. Also, they have made it more difficult for law enforcement to trace the victims and identify the traffickers. Sadly, more than fifty percent of the victims of sex trafficking are children, many of which are exploited through the Internet.

The first step for preventing and fighting human trafficking is to identify the traffickers. The primary goal of this study is to identify potential organized sex trafficking networks in Louisiana by analyzing the ads posted online in Louisiana and its five neighboring states. The secondary goal of this study is to examine the possibility of using authorship attribution techniques (in addition to phone numbers and ad IDs) to group together the online advertisements that may have been posted by the same entity.

The data used in this study was collected from the website Backpage for a time period of ten months. After cleaning the data set, we were left with 123,436 ads from 47 cities in the specified area. Through the application of network analysis, we found many entities that are potentially such networks, all of which posted a large number of ads with many phone numbers in different cities. Also, we identified the time period that each phone number was used in and the cities and states that each entity posted ads for, which shows how these entities moved around between different cities and states.

The four supervised machine learning methods that we used to classify the collected advertisements are Support Vector Machines (SVMs), the Naïve Bayesian classifier, Logistic Regression, and Neural Networks. We calculated 40 accuracy rates, 35 of which were over 90% for classifying any number of ads per entity, as long as each entity (or author) posted more than 10 ads.

Date

6-4-2020

Committee Chair

Van Scotter, James R.

Share

COinS