Using Data Science for Social Good: Fighting Human Trafficking
1 like530 views
Vincent Emanuele discusses the role of data science in combating human trafficking, particularly through the development of community-based initiatives like Anidata. The document highlights the alarming statistics on commercial sexual exploitation of children and describes a series of hackathons aimed at creating data-driven solutions to aid law enforcement in tracking traffickers. Key outcomes included the creation of tools like the 'pimp profiler' to gather and analyze data on potential traffickers to improve response times and increase convictions.
Using Data Science for Social Good: Fighting Human Trafficking
- 1. Using Data Science for Social Good: Fighting Human Trafficking Vincent A. Emanuele II, Ph.D President/Co-founder, Anidata info@anidata.org August 31, 2016
- 2. Why I’m here Here to tell you a story ● A group of aspiring and practicing data scientists ● Found an important problem to work on together and in collaboration with professionals in the community (human trafficking) ● Had some successes and failures along the way, with a ton of personal and career growth We are a group of people who are JUST LIKE YOU
- 3. Professional Bio ● PhD in Electrical and Computer Engineering from Georgia Tech(Signal Processing and Machine Learning) (2010) ● CDC Visiting Scientist (2006 - 2013) ● First data scientist at Wellcentive. Founded Data Quality, Data Governance, and Data Science Teams (2013 - 2016) ● Co-founder of Anidata (2016) ● Founder of Zylinium Research (2016)
- 4. Hello, I’m Vince, and I’m a “Refugee PhD”
- 5. Guiding Principle #1: Mentorship Make sure the professionals (data scientists) who come behind you end up better data scientists that you were, faster than you were able to get there
- 6. Guiding Principle #2: Community Improvement Leave the community you live in better than how you found it
- 7. And this is how Anidata came to be There are many people with both the Aptitude and Motivation to become data scientists They are missing 1) good data science mentors 2) exposure and experience working on real projects This is how Anidata plans to serve the community
- 8. The Human Trafficking Problem Commercial Sexual Exploitation of Children Information provided by: Irina Khasin, JD Chief Senior Assistant District Attorney Human Trafficking Unit, Fulton County District Attorney’s Office
- 9. Commercial Sexual Exploitation of Children (CSEC) ● How many children run away from home annually in the US? ○ 1.6 million ● In 2014, the National Center for Missing & Exploited Children estimate that 1 in 6 endangered runaways reported to them were victims of sex trafficking ● For teen runaways, what percent will become victims of CSEC? ○ 1 in 3 teens will become a CSEC victim within 48 hours of leaving home and becoming homeless (“survival sex”) ● What is the average age of entry into the commercial sex industry in the US? ○ 12-14 years old ● In 2014, the National Human Trafficking Resources Center hotline, operated by Polaris, received reports of 3,598 sex trafficking cases inside the United States
- 11. The Truth about CSEC ● Victims don’t choose prostitution ● Victims are young and vulnerable children ● Average age of entry is 12-14 years old ● Overwhelming number are runaways ● 70-90% have been sexually abused ● Most come from troubled homes ○ Abuse, neglect, parental crimes, and addiction ● Truancy and school problems are common
- 12. How does CSEC happen? ● Traditional ○ The track: Fulton Industrial/Old National ○ Truckstops ○ Massage parlors/strip clubs ○ Residential Brothels ● Modern Trend ○ Backpage ○ Craigslist ○ Escort services ● A digital data trail is a logical target for a data science-centric approach to this problem!
- 13. Building a Data Infrastructure to Fight Human Trafficking
- 14. Getting and organizing the data This is often said to be 80% of a data scientist’s work and it can be messy Approach ● Target adult web endpoints that are known to be channels for human trafficking ● Write python code to scrape advertisements from these sites ○ We choose NOT to store or process any image data ● Store in a database ● Anonymize our data collection by using Tor and make requests look “human-like” ● Focus on the southeast region, with Atlanta as a central anchor but extending as far as Nashville, TN After running for 2.5 months, we collected a database of approximately 160,000 ads
- 15. Database Schema: Human Trafficking Digital Signatures
- 16. Anidata Hackathon 1.0 Pitch Your Product Idea and Find a Product/Market Fit March 5, 2016
- 17. Anidata 1.0 Human Trafficking Hackathon Purpose: The purpose of this hackathon is to develop one or more prototypes and/or data-focused product concepts that could lead to a measurable impact on preventing, prosecuting, investigating, or otherwise aiding the fight against human trafficking within the next 12 months. One day hackathon on Saturday, March 5 from 7:30AM to 7:30PM Free food and caffeine all day Invite-only to Anidata members Judging: This is a non-competitive hackathon. We will not be judging “winners”. However, all teams are required to give a 7 minute pitch presentation to the “pitch panel”, which will consist of recognized experts from the human trafficking/sex crimes legal, investigative, and research community. The pitch panel will give invaluable feedback on the value of the product concepts demonstrated. We will strictly enforce the time limit for the pitch presentations.
- 18. Collaborations Data scientists ● We had 14 data scientists and aspiring data scientists Law enforcement ● We had 12 volunteers from the District Attorney’s office ○ Prosecutors, investigators, law-enforcement, victim interview specialists
- 19. Hackathon Schedule Time Task Primary Contact 7:30:00 AM Check-in: Breakfast & Meet and Greet Vince/Irina/Bob/Dan 8:00:00 AM Introductions Vince 8:15:00 AM Walk through of the data and existing infrastructure (informal w/o slides) Dan 8:45:00 AM Team and task formation Vince/Bob 9:00:00 AM Start Hacking Vince 10:00:00 AM Mentor Office Hours I Vince 11:45:00 AM Lunch & Team Status Updates (informal) Vince 12:15:00 PM More Hacking Vince 4:00:00 PM Mentor Office Hours II Vince 4:30:00 PM Final push and polish hacking Vince 6:00:00 PM Product Pitches (7m per team) Vince 6:30:00 PM Pitch Panel Recess (share notes, discuss, prepare feedback) Vince 7:00:00 PM Pitch Panel Review of Product Pitches Pitch Panel/Irina 7:30:00 PM Concluding Remarks Vince/Irina
- 20. Results of Hackathon 1 We had four teams pitch product concepts, with some ideas partially implemented and some demos Pitch panel of Human Trafficking experts gave us INVALUABLE feedback to shape the product directions Recurring theme across teams: Connecting the dots to understand underlying trafficker as fast as possible is important. We identified a product-market fit, and we abstracted the real data science/business problem the DA faced as follows….
- 23. The Product-Market Fit Concept: The Pimp Profiler
- 24. Our star characters in this story Meet Mary, our human trafficking victim Mary ran away from an abusive home at the age of 16 Within 48 hours, she was “recruited” by Zed Mary is not alone. One out of three juvenile runaways are propositioned by a trafficker or purchaser within 48 hours
- 25. Our star characters in this story Meet Alice, our Assistant District Attorney in the Fulton County District Attorney’s Office She was called by APD to interview Mary at a placement after she had been recovered
- 26. Our star characters in this story Meet Bob, the investigator helping Alice collect evidence during her interview with Mary If Alice and Bob can work fast enough, they can identify Zed and get a warrant for his arrest Most victims like Mary run away from placement within a week This initial investigation is CRITICAL
- 27. We call this window of opportunity “The Golden Week” Mary put in a placement Mary arrested by APD APD calls Alice to ask for help Alice and Bob Interview Window Alice and Bob Interview Window Mary runs away from placement
- 28. Interview Alice interviews Mary to discover Zed’s real identity: Alias Phone Number
- 29. Alice Creates a Profile with the Pimp Profiler Legal Name: XXXXXXXX Alias: Zed Email IDs: XXXXXXX Phone: (828) 280-2330 (This is a stolen phone)
- 30. Pimp Profiler Fetches Relevant E-mails Legal Name: XXXXXXXX Alias: Zed Email IDs: 'heavensent.trinity@yahoo.com', 'islandgirl.sapphire@yahoo.com', 'islandgirl.sapphire@gmail.com', 'heavensent_trinity@yahoo.com' Phone: (828) 280-2330
- 31. Pimp Profiler Finds Associated Phone Numbers Legal Name: XXXXXXXX Alias: Zed Email IDs: 'heavensent.trinity@yahoo.com', 'islandgirl.sapphire@yahoo.com', 'islandgirl.sapphire@gmail.com', 'heavensent_trinity@yahoo.com' Phone: (828) 280-2330 Alternate Phone Discovered: (828) 242-9969 (Zed’s personal phone number)
- 32. Zed’s Identity Discovered Legal Name: Dan Michaels Alias: Zed Email IDs: 'heavensent.trinity@yahoo.com', 'islandgirl.sapphire@yahoo.com', 'islandgirl.sapphire@gmail.com', 'heavensent_trinity@yahoo.com' Phone: (828) 280-2330 Alternate Phone Discovered: (828) 242-9969 (Zed’s personal phone number)
- 33. Victim Identifies Dan in a Line-up
- 34. Justice Served
- 35. Projected Measurable Impact on Law Enforcement ● Reduce time between first victim interview to arrest of perp ● Increased number of convictions per year ● Increased number of indictments per year
- 36. Anidata Hackathon 1.1: Building a Minimum Viable Demonstration Hackweek
- 37. Stages of Innovation MVD: “Minimum Viable Demonstration” ● Not quite a prototype, more than a mockup, and something that can be shown in a simple demo
- 38. Anidata 1.1 Hackweek MVD Plan ● Goals ○ Create polished demo ○ Gain full stack skills ○ Have fun collaboratively building a “product” ○ Tighten up the cool work we did in Anidata 1.0 ● What we hoped to create ○ A mobile-responsive web UI that allows one to query the dataset and browse returned results ○ A backend service that allows the web frontend to query the database for results ○ Clean algorithm code that can be run on scraped data ■ Phone number cleaning ■ Email type combing ■ Phone-email-UID entity resolution (people graphs) ○ A mechanism to run everything and turn on the web service
- 39. Hackweek MVD Architecture Backpage DBScraper Frontend ETL
- 40. “Connect-the-dots” Framework: Graphy Theory The information attached to distinct trafficking entities can be modeled using graph theory Connected subgraphs represent all information we know about a single underlying trafficker
- 41. “Connect-the-dots” Framework: Graphy Theory
- 42. Web App Prototype: The Stack node.js angular.js Ruby on rails Python networkX Postgresql
- 43. Conclusions ● Hackathons are a great way to build the data science community ● Hackathons with partners in the community are a great way to raise awareness of how data science can solve business problems ○ Caveat: If you focus on the business problem ○ This helps create jobs and a local economy for data science ● Event planning is time consuming ● We are not that different from YOU. Just a group of motivated individuals with a common purpose.
- 44. Questions? Vincent Emanuele - Add me on LinkedIn! Contact Info: info@anidata.org vincent.emanuele@anidata.org