This is a Kaggel dataset for text analysis. The original dataset has some toxic comments of some types(like obscene, threat, insult). Although I modified it a little bit. I just lebelled them as 1 i.e. toxic. So my model is a binary classifier. I used Tenserflow, a very popular Deeplearning open source tools to make this model. It gave an accuracy of 95% on a training of 3 epoch with 2 conv(1D) and a Dense layer.
I provided the training and test dataset with the output saved word embedding model(tsv). You can download the repo and the instruction for running it is given there. You also can take the reference of the pdf. Related theory is also proivded there.