Using scikit-learn to find bullies – Towards Data Science – Medium

To find a small dataset to play with, I found a Kaggle competition from way back when: Detecting Insults in Social Commentary. The training set has 3000 datapoints, which is 100 times smaller than the last natural language processing challenge I tackled. The challenge: Identify whether a comment would be considered insulting to another participant in the conversation. The comments were taken from commenting sites, message boards etc, and were provided as csv files. As the above examples show, it was necessary to clean the sentences.