This repository contains experiments with convolutional neural networks (CNNs) for the detection of toxic online comments. The experiments are intended to help answering the following questions:
- How suitable are different CNN architectures for the toxic comment classification task?
- Which preprocessing steps are beneficial for this type of problem?
- Which patterns do CNNs learn and which features are important for the classification when being trained on this data?
- Do CNNs work equally well when being applied to data coming from different sources?