MS Thesis
Published:
The code behind my master’s thesis at METU: a pipeline that detects cyber-security events in Turkish, a language with no labelled security corpus to train on.
It collects the raw material first. Selenium and BeautifulSoup scrapers pull Turkish newspaper archives, and a separate collector reads the Twitter stream, which is where an incident usually surfaces before it surfaces officially. On top of that sits the detection work: a keyword vector learned from an incident whose answer is already known, normalisation for Turkish morphology (one stem becomes dozens of surface forms, and a bag of words treats them as unrelated tokens), and a per-entity anomaly test rather than a sentiment classifier.
The method and its measured results are written up in the thesis and the ICISSP paper: Automatic Detection of Cyber Security Events from Turkish Twitter Stream and Turkish Newspaper Data. There is also an animated walk through the argument in the lab.