Machine Learning to Detect Illegal Online Sales of Prescription Opioids

on YouTube ↗18m

" summary

Pipeline ingests drug keyword streams via Twitter streaming API and collects ~1M tweets over months, applies a 2013 bi-term topic model to surface a reproducible topic whose top words include "best free pills online by" flagging illicit pharmacy posts, manually labels those clusters with about 90% precision, engineers metadata features (URL presence, hashtag use, high friends to followers ratio) and trains a logistic regression that achieves high AUC/precision/F1 for detecting illegal opioid sales.

AI-written summary · about

" chapters (5)

1. 0:00Research motivation
2. 1:43The impact of illicit drugs
3. 6:37Data collection pipeline
4. 10:39Topic modeling for detection
5. 14:03Feature engineering and models

" related talks

• 42mGraph Theory Takes on the Opioid EpidemicMark Quinsland · Data Con LA 2019
• 32mAnalyzing social media for suicide risk using natural language processingPhilip Resnik · AWS Machine Learning Summit 2021
• 21mMachine Learning and Cyber Security - Detecting malicious URLs in the haystackVeatrissa L. · PyCon AU 2019
• 41mUsing Machine Learning to Help with Daily WorkflowsDanny Teng, Neha Bhomia · DrupalCamp NJ 2020
• 06mOpiTrack: A Wearable-based Clinical Opioid Use Tracker with Temporal Convolutional Attention ...Bhanu Teja Gullapalli, Stephanie Carreiro, Brittany P Chapman, Deepak Ganesan, Jan Sjoquist, Tauhidur Rahman · UbiComp 2021
• 24mHunting Criminals with Hybrid AnalyticsDavid Talby · Data By the Bay 2016
• 42mDetecting novel anomalies in TwitterDelia Rusu, Mattia Boni Sforza · PyData London 2016
• 26mExtracting Medical Attributes from Clinical TrialsSanghamitra Deb · Data By the Bay 2016