Check generality

[timm]

• get text data

Repeat 20 times

1. separate 90:10
2. text ==> topics
3. topics ==> classes (e.g. buggy)
   • needs some param tuning (e.g. words per topic)
4. new data sets:
   • top top words in each topics + LDA
5. classifying with SVM

[timm]

does this work for technical debt and stack overflow

[timm]

• need to justify your baseline