Seminar S’21 – Combating Human Trafficking via Automatic Intelligence Collection, Validation and Fusion
- Post by: Bahareh Arghavani
- November 29, 2022
- Comments off
Presenter: Bibek Upadhayay (SAIL Lab)
Time: Friday 2/12 , 3pm – 4pm ET
Recording: https://youtu.be/q9iFGlSfFhk
Abstract:
One of the major social problems in the world is human trafficking that accounts for 24.9 million people being victims yearly which includes 49% of women and 30% of children. The problem of human trafficking can be tackled in two aspects, the first aspect is to create strong law and policies from government for the human trafficking. The second aspect is to perform community level work to mitigate the root cause of human trafficking. The challenge we face here is that to find the root cause at community level we lack sufficient and timely data. The initiation from LoveJustice and UNH SAIL Lab, we are trying to remove the bottleneck of needing human analysis to produce actionable intelligence from unstructured and semi-structured data plus the overhead of data flow and data integration in analytic pipelines. In our first phase we have planned to develop a multi-source OSINT collector that crawl the focused news articles and extract and filter the text and populate the Case Information Form fields for top priority fields. We have divided the overall works into two parts, the first part is News Collector and filter module that contains the news crawler that makes use of the language models to create search phrases to crawl more relevant news. And the second part is Information extractor that contains the deep learning NLP model and algorithm to that finds the fields in the case information form such as victim name, suspect, meta data of victims and suspect such as age, gender, education, address, and nationality. Our model and algorithm yielded the accuracy of 73% on identifying victims, 65% on identifying gender of victim 59% identifying the age of victim. The overall model also proves to work at a cheaper rate of $0.10 for processing one article which is nominal rate compared to human cost.
Bio:
Bibek Upadhayay is a graduate student in the MS Computer Science, and a research assistant at SAIL-Lab. He is also the recipient of the 2020-2021 TCoE Graduate Research Fellowship. Bibek’s research is on machine learning applications to fake news detection in online social media. His work on this project has so far led to the curation of a new dataset of Fake Claims, called “Sentimental Liar“, as well as a novel deep learning approach to the classification of fake news in a short text and paper on this ‘Sentimental LIAR: Extended Corpus and Deep Learning Models for Fake Claim Classification’ has been accepted and presented on IEEE ISI 2020 conference. Bibek has also been working on Combatting Human Trafficking and Cyber Threat Intelligence project.