Reading Thieves' Cant: Automatically Identifying & Understanding Dark Jargons...
summary
An automated system identifies words whose contexts differ between legitimate and underground corpora by training a dual-input word embedding with a shared output layer, then uses projection learning to infer each jargon’s hypernym. The implementation, Cantreader, runs over one million traces from four forums and reports 3,462 dark jargons with 2,491 novel hits, finds jargons in about 25% of traces, and evaluates at precision 0.91 recall 0.77.
AI-written summary · about
chapters (8)
- 0:00Dark jargons introduction
- 2:03Detecting dark jargons
- 4:36Word2Vec semantic vectors
- 8:58CSSM for cross-corpus
- 15:45CSSM for detection
- 16:50Hypernyms for understanding
- 18:57System performance evaluation
- 20:53Q&A
related talks
03mJeremiah O'Connor
21mMaking Sense of Unstructured Threat Data
26mNLP for security log analysis : Learning to crawl before you run
29mDetecting anomalous sequences using text processing methods
50mDeep Impact: Recognizing Unknown Malicious Activities from Zero Knowledge
17mDetecting anomalous sequences using text processing methods
51mFrom Text to Flaws: vulnerabilities in applications with Generative AI and LLMs
40mIRonMAN: InterpRetable Incident Inspector Based ON Large-Scale Language Model and Association miNing