Detecting toxic comments with multi-task Deep Learning

The internet is a bright place, made dark by internet trolls. To help with this issue, a recent Kaggle competition has provided a large number of internet comments, labelled with whether or not they're toxic. The ultimate goal of this competition is to build a model that can detect (and possibly sensor) these toxic comments. While I hope to be an altruistic person, I'm actually more interested in using the free, large, and hand-labeled text data set to compare LSTM powered architectures and deep learning heuristics. So, I guess I get to hunt trolls while providing a casestudy in text modeling.