Hacker News Reader

grouping dont always yield better results, and I think it would probably do pretty poorly in this case because there are few observations.

Random Forest won in the samples tried, but I wager a support vector machine with a histogram kernel would do fantastic.

We could always use pre-trained word embeddings, the few observations won't matter then.

this is very industry specific, finding them that are meaningful for the data at hand seems like a small chance.

edit: im not saying dont try it, i certainly would! lets look at the data on github? maybe we could have a wack at the data