Predicting English keywords
from Java Bytecodes
using Machine Learning

Short summary

Machine learning on 330k de-compiled Java methods from the Debian archive plus their corresponding Java Doc textual comments. Input are bytecodes, output are English keywords. If you wanted an excuse to play with machine learning, it hardly gets better (and more challenging) than this. Work in progress.

Downloads and demo

• Upload a .class and see a demo and sample output (code is in github).
• Slides from Pablo's talk at the RALI-OLST Seminar
• Slides from Pablo's RECON2012 talk (pdf).
• Training data (50Mb, 7zip compressed).
• Trained models, first version (34Mb, 7zip compressed, 1.3G decompressed).

Contact:

• Follow @kwd4bytecodes
• Contact the author.

 

---

Copyright (C) 2012 Pablo Ariel Duboue. Data made available under CC-BY license.