Show HN: Cyber-forensics for public PCs using ML to detect hate speech, threats

4 points

8 years ago

This is a digital forensics project I've been working on called OLAF which is designed to provide forensic evidence in investigations of using library or other public-facing PCs to send hate speech, violent threats, "swatting", IP exfiltration or infiltration, or to commit any other type of computer-enabled crimes or malfeasance. The project uses OCR, NLP, and computer vision to create high level logs of the PC's interactive usage that can be used in investigation or detection of dangerous or criminal activity.

Demo vid 1: https://youtu.be/F6k2PC-7WQw

Demo vid 2: https://youtu.be/-N2Qftl26Mw

Project repository: https://github.com/allisterb/OLAF

The raw data OLAF uses is logged in disparate places in one form or another (like in your PC or ISP or web server logs) and OLAF just tries to do real-time analysis of the available data in one place. No PII is ever intentionally logged and it is up to the org to link PC user accounts and identities to people.

Of course privacy is a big concern here and the organizations using a tool like OLAF have to walk the line between protecting the privacy of their users and adhering to the relevant laws vs. being able to quickly detect and investigate potentially serious threats to people's safety. Guidelines like those from NISO (https://www.niso.org/publications/privacy-principles) should be adhered to as much as possible.

Libraries currently used:

Tesseract.net: https://github.com/tvncosine/tesseract.net

VADERSharp: https://github.com/codingupastorm/vadersharp

Accord.NET: https://github.com/Accord-net/framework/

Azure Cognitive Services Computer Vision API

Azure Cognitive Services Text Analytics API

Let me know what you guys think and any comments or suggestions (or criticism) you may have.